In today’s column, I provide seven keystone predictions about what will occur for generative AI in the upcoming new year of 2024.

Four of the predictions focus on advances in AI technology that will alter the landscape of how generative AI is devised, deployed, and utilized. Three of the predictions underscore the business and societal considerations of how generative AI will be assessed, acclaimed or denounced, and potentially regulated or governed. That is seven all told. Seven is said to be a lucky number. Yes, in that sense, my predictions turn on both a dollop of skill and a dash of luck.

Readers might recall that last year I made several predictions about what would happen with generative AI for 2023, see the link here.

Last year, I identified five keystone predictions, which I then further detailed into five sub-predictions for each major instance (thus, a total of 5×5 = 25 predictions). By and large, the predictions turned out to be on-target and the essence of each came to approximate fruition. This is not to suggest that I had a Nostradamus magical touch. I mindfully sought to derive a practical set of predictions within the realm of reasonable possibility and based my predictions on a deep grounding of what was being envisioned and percolating inside a wide array of AI research labs and AI startups that I am involved with. Plus, my AI consulting work, AI conference speaking engagements, and extensive activity on AI standards committees provide a rich source of AI insider alertness.

This time I am undertaking an akin rationalistic approach and have come up with seven major predictions about generative AI for 2024. Thus, I went from five last year to seven for this upcoming year. I decided not to sub-divide the seven since the feedback last year was that there was a proverbial forest for the tree’s preference in how my predictions should best be conveyed (thanks for the handy feedback).

Seven is a good number and in this case, suitably covers what I consider to be the most momentous or jackpot of upcoming changes and advances in generative AI.

Another reason to relish the number seven is that per the now classic and famous work of cognition researcher George Miller, author of “The Magic Number Seven, Plus Or Minus Two: Some Limits On Our Capacity For Processing Information” (The Psychological Review, March 1956), he said this about the amazing number seven: “And finally, what about the magical number seven? What about the seven wonders of the world, the seven seas, the seven deadly sins, the seven daughters of Atlas in Pleiades, the seven ages of man, the seven levels of hell, the seven primary colors, the seven notes of the musical scale, and the seven days of the week? What about the seven-point rating scale, the seven categories for absolute judgment, the seven objects in the span of attention, and the seven digits in the span of immediate memory?”

I aim to try and showcase the seven most significant predictions about generative AI for the year 2024. I hope they will be of use to you. Fortunes can be made, and likewise, fortunes can be lost, all based on what emerges as generative AI continues to advance.

Make your bets wisely.

My Seven Predictions About Generative AI For 2024

Let’s go ahead and jump into the predictions. I’ll start with the four that are related to AI technological advances. After covering each of those, I’ll proceed to indicate the three that are more so of a business or societal impact nature.

The seven consist of these categorical areas of generative AI:

• (1) Generative AI Multi-Modal – Makes majestic mainstay manifestly moneymaker
• (2) Generative AI Compacted – Capability categorically cautiously convincingly considered
• (3) Generative AI E-Wearables – Enter evocatively everywhere eventually earnestly
• (4) Generative AI Functionalities – Fervently favor focus for fruitfulness
• (5) Generative AI GPUs – Golden gadgetry gaps gaspingly glaringly
• (6) Generative AI Courts – Confusion copyrights consternation cajoling congress
• (7) Generative AI Deepfakes – Disrupt democracy devilishly dastardly disturbingly

You might notice that I opted to add a bit of whimsy to the seven by providing a catchy shorthand description. That’s the extent of my poetic talents. Perhaps at least the zesty descriptions whet your appetite and get you mentally energized for my rundown of the seven. Hope so.

At the end of each elaboration, I provide links to my column coverage that pertains to the topics addressed.

Here we go.

(1) Generative AI Multi-Modal Makes Majestic Mainstay Manifestly Moneymaker

In 2024, multi-modal generative AI is finally coming to fruition and will be a dominant focus throughout the new year.

Here’s the deal.

Generative AI has so far been principally a single-mode or mono-modal affair. The early days concentrated on the text-to-image or text-to-art mode. You could enter some text and the generative AI would produce an image or artwork for you. People enjoyed doing this. The commercial prospects at the time were relatively low. It was mainly a fun thing to do and be able to gape in amazement at what the generative AI could pictorially derive.

Next, along came ChatGPT and the rise of text-to-text as a mode of generative AI activity.

You enter text and you get text in response to your text. Sometimes I refer to this as text-to-essay so that I can emphasize that generative AI does more than just perhaps simplistically echo your text back to you. A crucial element of this text-to-text or text-to-essay is that you can do this repeatedly in an interactive manner. Some users treated generative AI as a one-and-done operation, namely, they entered a piece of text, got an answer, and logged out. The true essence of using generative AI is that you can carry on a seemingly fluent dialogue with the AI. Your mindset about how to use generative AI is vital to what you can gain from using the AI.

Toward the latter part of 2023, we began to see the wider emergence of a limited form of multi-modal generative AI, wherein a generative AI app might allow for two types of modes, such as allowing you to do text-to-text and also do text-to-images or text-to-art. This was a baby step forward. You know how it is, sometimes you need to crawl before you can walk, and walk before you can run.

We are now on the verge of entering widely into the use of multi-modal generative AI more robustly and extensively.

Generative AI will provide numerous text-to-X modes, including:

• Text-to-text
• Text-to-images
• Text-to-art
• Text-to-audio
• Text-to-video
• Text-to-other

Perhaps the most vaunted and exciting will be the text-to-video. The idea is that you can describe something of interest in text and the generative AI will produce a video corresponding to your description. This is already beginning to happen but on a very limited basis such that the video is merely a short grainy clip and only tangentially transforms your text into an engaging and salient video.

The next step for text-to-video will be to produce a professional-quality full-length video that embellishes and vividly showcases whatever text you entered. Imagine that you write a script for a movie or TV show and want to have a video generated based on that script. Furthermore, imagine that the video looks fully derived including a convincing visual portrayal of characters, places, actions, and the like.

We are heading toward enabling just about anyone to readily and at low cost make a film or video that they can then show to the world and potentially make money from. Some believe that this is yet another step in the “democratization” of filmmaking and that the major studios and entertainment-producing firms are going to be dramatically disrupted.

Anyway, I want to get back to the multi-modal considerations.

We can also go in the opposite direction of text-to-X with the modes of X-to-text:

• Image-to-text
• Art-to-text
• Audio-to-text
• Video-to-text
• Other-to-text

Let’s again consider the video instance. Think about the video-to-text mode. You can feed a video into generative AI and get the AI to indicate what the video consists of. This can be done today but the text generated that depicts the video is very stilted and doesn’t do a notable job of explaining or describing the video. The idea is that an entire text essay of a flowing nature would be generated. Envision for example a well-written book or novel that was written by generative AI as based on a film or TV show.

That’s the kind of amazement we will start to see in 2024.

Those overarching modes that I’ve mentioned about generative AI are generally known respectively as X-to-text and text-to-X. Note that text is part of the equation, either as the input or the output (or both sides of the equation).

Another angle is the fully generalized X-to-X and goes beyond the text as a part of the equation, including:

• Image-to-video
• Video-to-image
• Audio-to-image
• Image-to-audio
• Other-to-other

We will also witness the X-to-X-to-X variations.

For example, I feed a video into generative AI and ask for a text description (video-to-text). Meanwhile, I write some text, add that to the text that I just got from generative AI, and feed that into a text-to-video mode and get an awesome new video produced. I might have done this to take a movie and make a follow-up movie that is the second in a series or a sequel. You can discern that this would have required the video-to-text, followed by text-to-text, followed by text-to-video.

The gist is that you can mix and match the multi-modals to your heart’s content.

The range then of multi-modal generative AI will be a combination of your choosing:

• Text-to-text
• Text-to-images
• Text-to-art
• Text-to-audio
• Text-to-video
• Text-to-other
• Image-to-text
• Art-to-text
• Audio-to-text
• Video-to-text
• Other-to-text
• Image-to-video
• Video-to-image
• Audio-to-image
• Image-to-audio
• Text-to-text followed by text-to-video
• Video-to-text followed by text-to-text followed by text-to-video
• Etc.

My prediction is that 2024 will be the advent of viable and useful multi-modal generative AI.

We will shift from the existing primitive versions to quite extensive versions. There are those people who are already starting to use multi-modal generative AI and will expand and deepen their use as the capability improves and becomes nearly commonplace. Mark my words, people will also think of ways to use multi-modal generative AI that nobody today is contemplating. Some of the uses will be for goodness, while the odds are that some of the ways will be for badness. The sky is the limit either way.

The multi-modal functionality is likely to attract more people toward using generative AI and we will go from the existing base to an even larger segment of the population. You can anticipate that those already comfortable with generative AI will undoubtedly readily embrace the multi-modal features as they are rolled out. New entrants will be aplenty.

Furthermore, the infusion of generative AI into other apps is going to skyrocket. Everyday apps that today have nothing to do with generative AI will be expanded to encompass generative AI. This is sensible because people are going to be uplifting their expectations about the use of everyday apps. They will expect everyday apps to provide a fluent interactive experience. The sensible way to get there will be to incorporate generative AI into those apps. A scramble to do so will ensue.

The year 2024 will be mainly initial experimentation and exploration as people begin to see how multi-modal generative AI works and then come up with ways to leverage the remarkable functionality. That being said, I would estimate that it won’t be until 2025 that bona fide solid uses of multi-modal generative AI are adopted on a widespread basis.

Say it with me, multi-modal generative AI is a big deal and 2024 will see the rise of multi-modal beyond our wildest dreams (well, maybe not quite that far, but you get the gist).

For additional details on this and related topics, see my coverage at the link here and the link here, just to mention a few.

(2) Generative AI Compacted Capability Categorically Cautiously Convincingly Considered

Go small or go home.

I realize that the usual saying is to go big or go home, but in the case of generative AI, the mantra for 2024 will be to go small or go home.

Here’s the deal.

Most of the generative AI apps require making a connection to a cloud server that is running the generative AI program. This makes sense due to the massive size of the best-in-class generative AI programs and their hungry need for computing resources. They reside in lots of beefed-up servers. You run the generative AI and doing so consumes highly expensive computational processing cycles. Depending upon which generative AI app you’ve decided to use, the cost to you for using those services might be free, while in other cases, you need to pay by transaction or some agreed metric for each use by the generative AI app that you are utilizing.

There are lots of downsides for the user or consumer to this arrangement.

First, as mentioned, you are potentially facing a big financial bill for the use of the pricey servers in the sky. Second, you must be able to find an available network connection including having reliable access to smoothly enable the generative AI to communicate with your edge device such as a smartphone, laptop, tablet, desktop, or other. Third, you expose yourself to potential privacy issues due to pushing your data and prompts up into the cloud where the AI maker or other third parties might be able to dig into it. Lots of concerns abound.

Ideally, you would greatly prefer to have the generative AI be entirely operating on your edge device. By having the generative AI on your smartphone or other device as a standalone you could then forego the need for a network connection, you would not have to pay for some faraway server to run the program, and you would be able to keep your data and prompts privately on your device. That is the happy face scenario.

The problem is that generative AI usually requires a massive amount of storage and a massive amount of computational processing to do its thing. Your smartphone or laptop won’t cut it. They are too underpowered and underequipped. That is the sad face scenario.

But those daunting challenges do not stop the eager pursuit of squeezing down generative AI to fit comfortably onto an edge device. Where there is a will, there is a way.

One aspect that comes to our rescue is that once a generative AI app has been initially data trained, the hefty weightlifting has been accomplished. Whereas devising generative AI soaks up a ton of disk space and computer processing, the runtime of merely executing or running the generative AI is tiny in comparison. Building generative AI from scratch almost always requires going big, while running does not necessarily need to be big and can be potentially cleverly devised to go pretty small.

Another aspect in our favor of the smallness aspiration is that AI research studies tend to suggest that you don’t usually make use of every nook and cranny of generative AI. There are large swaths that you are unlikely to touch upon. Furthermore, in a technically astute manner, you can potentially compress the generative AI system’s massive data structure into a much smaller size, including exploiting sparse areas of the artificial neural network (ANN) that do not provide any added value per se.

All in all, there is going to be an emergence in 2024 of having generative AI that is entirely standalone and can reasonably work on an edge device like a beefy smartphone or laptop. We already have this somewhat though the versions of generative AI that do so are typically a far cry from their bigger brethren. They aren’t as extensive and aren’t as fluent. Nonetheless, they are proof of concept that keeps everyone in the game of aiming to compact generative AI.

The impact will be significant once these reach a more satisfactory level of capability. Users will relish having an entire generative AI that resides on the smartphone and will work anywhere and at any time. No online connections are needed. No servers are required. And you’ll be able to embrace privacy that allows you to keep your data and prompts secured in your own safe space.

Companies are going to love this too. Their costs to use and deploy generative AI will be reduced. AI makers are going to begrudgingly like this, they will do so with a teeth-grinding sense of angst. Why so? Several reasons. First, they won’t be able to garner financial profit from your use of their servers. Second, they won’t be able to readily control what you do with their generative AI. Third, they won’t easily be able to reuse your data and prompts, which many of the AI makers do currently to enhance the data training of their wares. Etc.

The AI makers will be faced with the potential touch choice of whether to continue to remain on the big iron or go with standalone compacted variations. Seems daunting. We should play a melancholy violin song for them.

Maybe not.

You see, I don’t perceive this as one of those make-or-break choices. The odds are that they will continue to use the big stuff for their truly state-of-the-art generative AI that will be so large and so computationally intensive that it won’t reasonably be squished into a small platform at this time (until the compaction methods catch up and can deal with the stepwise increase in size; it is an ongoing cat and mouse gambit). Meanwhile, they will presumably be brainy enough to realize they need to take their regular or legacy-sized generative AI and squeeze more life out of it by compactly making those versions available.

They can have their cake and eat it too.

The bottom line is that smallness is the overarching goal. You can freely anticipate that 2024 will be the era of compacted generative AI that flies like an eagle on your smartphone, smartwatch, or other edge device. Get ready or go home.

For additional details on this and related topics, see my coverage at the link here and the link here, just to mention a few.

(3) Generative AI E-Wearables Enter Evocatively Everywhere Eventually Earne

You are what you wear.

Do you think that is a valid truism?

Well, if so, in 2024 you will be able to test your resolve on this often-exhorted adage.

Here’s the deal.

We are finally going to have in the public marketplace some relatively sophisticated e-wearables that will be worthy of your consideration. You will be able to wear a pendant, pin, necklace, ring, or other accessory that will be computer-jampacked with lots of nifty hardware features and will excitingly be making use of generative AI.

Many of these have already been preannounced. Few of them are yet readily available. The ones that were already available in the marketplace were typically lacking in how well they worked. The hardware features were minimal. The weight of the device was questionable. You had to be a diehard early adopter of new tech to want to pay through the nose to get relatively limited capabilities and something that was nearly unbearable to wear on your body for any sustained length of time.

The ease of use was also dismal.

That’s something that can be solved via modern-day generative AI. Let’s remember too that the generative AI in 2024 is going to be multi-modal. Tie together all these facets and you have the makings of e-wearables that people might find attractive for wearing on their persons.

Allow me to elaborate.

You opt to use a pendant or pin that is a modern e-wearable and incorporates generative AI. The pendant or pin also includes a plethora of hardware such as a tiny microphone, a tiny speaker, a tiny camera, and other stuff. When you are wearing the device, the microphone is listening for whatever you might perchance say. The tiny camera sends pictures or videos to the generative AI. Your utterances and what is being seen are sent to generative AI. Generative AI uses its natural language processing (NLP) fluency capabilities to try and discern what you are saying and what you are looking at.

You might be considering buying a new car and staring longingly at a car that is the type you want to get. You say aloud to no one in particular that you wonder how much that car costs. The open question is captured via the audio and sent to generative AI, which looks up the brand of the car as scanned from the camera image and responds by sending audio to the tiny microphone that then tells you the retail price for the vehicle.

Voila, you are walking around with generative AI readily at your whim.

Multi-modal generative AI at its finest.

The first half of 2024 will witness a slew of these hitting the shelves or being ordered and delivered via online stores. By the second half of 2024, we will either see an immense throng of people rushing to get them, or it could be a big fizzle and people decide these new mobile portable wearable electronics aren’t worth their weight in gold.

Some believe that we will go through a series of classical tech cycles. The generative AI-empowered e-wearables will have an initial wave that is not quite fully figured out. Bugs will exist. People will get upset. Pendants or pins or rings won’t be worn after a few days or weeks of tryout. They will end up at the flea market for a rock-bottom bargain price.

Call that version 1.0 of the generative AI e-wearables. On the heels of the 1.0 will come the 2.0. The generative e-wearables in the 2.0 era will be markedly better than the first incarnation. Whether we can go from 1.0 to 2.0 in just the year 2024 is a seeming stretch of the imagination. Time will tell.

Another point comes to mind that is relevant here. Recall that I mentioned that most of today’s high-end generative AI needs to use servers and be accessed via a network connection. The early round of e-wearables will be in that same boat. They will have to maintain an online connection for the generative AI to be active. I think you can see why my other point about compacting generative AI for use on edge devices comes into play here, namely that the e-wearables makers would certainly like to make the generative AI standalone and reside solely on an e-wearable device.

There loom some gloomy issues on the horizon.

Will the use of an always-on or always-available generative AI be a potential privacy intrusion on the person wearing the device? What about others that are near the person and thus the device is potentially recording them and parsing their utterances too? You can expect that tremendous societal AI ethics and AI law issues will push lawmakers and regulators to take renewed action about AI and the privacy and security of the public at large.

As a side note, I have ordered some of these new e-wearables that use generative AI, which I’ll be reviewing in my columns, so be on alert for that coverage.

I have a quick question for you. If I choose to wear these space-age e-wearables all of the time, what does that say about me (i.e., “You are what you wear”)? Then again, please don’t answer that question.

For additional details on this and related topics, see my coverage at the link here and the link here, just to mention a few.

(4) Generative AI Functionalities Fervently Favor Focus For Fruitfulness

Expertise is not a dime a dozen.

When you consult with a medical doctor, a lawyer, or just about any professional, the odds are that you are tapping into their personal expertise. Those of you who were around in the days of expert systems, knowledge-based systems, rules-based systems, and the like will remember that those were the times when everyone was aiming to infuse expertise into AI systems.

We are coming back to the future by wanting to do the same but now do so inside of generative AI.

Here’s the deal.

Generative AI is usually data trained on a wide swath of content on the Internet. You might say that generative AI is breadth-oriented rather than depth-oriented. I refer to this as generic generative AI that generally has been data-trained on a lot of topics. A master of none.

Meanwhile, we keep seeing people trying to apply generative AI to things like providing medical advice on par with a medical doctor or providing legal advice on par with a lawyer, etc. Not a good idea. Let me say that differently, it is a really bad idea. Most of the AI makers clearly stipulate in their licensing agreements that you aren’t to use their generative AI for such matters. Stick with the generics is what you are supposed to be doing.

I would wager few people realize this, nor do they care. They proceed anyway to try and put a square peg into a round hole. People will do as they want to do.

The problem of course is that the generic generative AI is insufficiently data-trained to provide viable answers to at times life-serious concerns. You will undoubtedly get answers. The answers could be completely off base. Sadly, people will tend to believe the answers because they “believe” whatever the generative AI tells them, especially if the portrayal oozes with confidence.

Like the old saying, if you can’t fight them, join them. For a whole bunch of really good reasons, there are many efforts underway to jack up generative AI and infuse expertise or domain specialties into generative AI. The idea is that you might go to a particular generative AI that has been for example infused with lots and lots of in-depth medical information. You are a bit more likely to get sound advice, though realize that none of this is going to be perfect.

How can we get generic generative to gain a foothold in specific functionalities or domains?

One approach would be to intentionally do this when you are first devising a generative AI app. At the get-go, you feed tons of data about a designated specialty. This then becomes part and parcel of the generative AI. Generative AI in this case is being trained on a wide swath of the general stuff on the Internet and at the same time being deep-dived into a chosen area of expertise.

Another approach asks what to do if the generic generative AI already exists. The horse is already out of the barn. You cannot wind back the clock and redo the generative AI from scratch. What are you to do when the generic generative AI doesn’t contain the desired expertise?

The answer would seem to be that you try to bring the generic generative AI up-to-speed, as best you can. Do additional data training after the fact. Acknowledge that you cannot turn back the clock. The road ahead consists of infusing new data into the generic generative AI and combining the new with the old.

A method for doing this that has become relatively popular is known as RAG (retrieval-augmented generation). I’ve described in detail how this works, see the link here. In brief, you collect together the data that you want to have the generic generative AI use. You pre-process the data and place a transformed version into a special file or set of files known as a vector database. Then, you instruct the generic generative AI to refer to the vector database when needed.

Generic generative AI can leverage a capability known as in-context modeling to be boosted via the vector database. When a user asks for something that the added expertise might be fruitfully utilized with, the generative AI accesses the vector database, brings in hopefully relevant content, mixes the content with the other aspects that the generative AI has in hand, and essentially adds this temporarily to the overall large language model (LLM) or modeling capability.

For short-term purposes and temporary determinations, this can be useful. Doing so extensively and in the long term is probably not going to be something of erstwhile value. Ultimately, we will need to either do the expertise infusing at the get-go or find some other means to include domain or functional expertise more seamlessly and persistently.

In 2024, you will likely be somewhat surprised to start seeing generative AI apps that claim to have “embodied” particular areas of expertise. The chances are they are using RAG or a similar confabulation. Be careful and make sure that you examine closely how the generic generative AI has been altered to include functional expertise. It might be done well, or it could be flimsy and highly suspect.

One thing that is nearly guaranteed in 2024 is that marketers will hype the heck out of their domain-steeped generative AI apps. You will need to figure out whether the claims are bona fide. I expect that lawmakers and regulators are going to get dragged into the fray. AI ethics and AI law will rise to the fore. It’s going to be a fun year or a chaotic one, or maybe both at the same time.

For additional details on this and related topics, see my coverage at the link here and the link here, just to mention a few.

(5) Generative AI GPU Golden Gadgetry Gaps Gaspingly Glaringly

I am now shifting from AI technological considerations to topics that are aligned with business and societal considerations. I trust that you noticed that even the AI technology topics had an element of business and societal undertones. That is the way of AI. AI is going to have technological facets and equally have business and societal ramifications. It is a package deal.

Shift gears.

I’d like to talk with you about graphical processing units (GPUs).

In the same manner that in the long ago past they used to say that a young up-and-comer ought to get into plastics (this was popularized in the famous movie The Graduate), you would nowadays be wise to tell someone to get into GPUs.

Here’s the deal.

When the AI field started devising large language models and generative AI, a cornerstone consisted of using artificial neural networks under the hood. These are somewhat inspirationally modeled after the idea of neural networks in the brain, though not at all the same and immensely crude in comparison.

The whole concoction of ANNs entails doing zillions upon zillions of mathematical calculations. A conventional computer processing unit (CPU) is not devised to expeditiously do calculations. The advent of video games led to the development of specialized chips known as graphical processing units aka GPUs that could achieve an enormous amount of very fast calculations. This is what helped video games to look good when displaying graphics and moving things on a computer screen.

Those of us in the AI field reasoned that we could borrow those GPUs and use the calculation prowess to aid in doing ANNs. It worked. Today, the reason that we have much of the modern-day generative AI is that when the initial data training is done and when you run the generative AI it relies upon a vast array of GPUs to do so.

The need for GPUs is growing and growing. The more people want to use generative AI, the more there is a need for GPUs. The manufacturers are overwhelmed with orders. Backlogs are aplenty. In a loosey-goosey sense, a GPU is worth more than its weight in gold.

A shortage of GPUs is upon us. This in turn could be said to be a potential barrier to making ongoing progress in generative AI. If you don’t have the core machinery available to devise and operate generative AI, you have gotten stuck up a proverbial creek without a paddle.

You can add to this issue that advances are being devised to make GPUs more powerful (faster, smaller, etc.), and more efficient in terms of not using up as much electricity as they otherwise might. There is a rising concern that each time you use generative AI, you are consuming environmental resources to do so. The aim is to find ways to make GPUs as efficient as possible and reduce the ecological profile involved in their usage.

Why should you care about GPUs?

Because the GPUs underly generative AI. If you want to have plenty of generative AI, you have to also (usually) have GPUs to do so. When the GPUs aren’t available, generative AI might be scarce or slowed down. GPUs are the engines that aid in keeping generative AI running along.

Let’s tie together several of the earlier predictions into this one.

Do you want to have multi-modal generative AI? Yes, I’m sure we all do. It takes GPUs to run the multi-modal generative AI. Without the GPUs, the multi-modal is mainly a pipedream.

Do you want to have e-wearables that can access online generative AI? Yes, that’s also on the desired list. Again, if GPUs aren’t available the generative AI that the e-wearables are accessing in the cloud will be scant or be especially costly (to the victors go the spoils).

Do you want generic generative AI that is augmented or infused with domain or functional specialties? Sure, that sounds amazing. But, if GPUs aren’t available or they are scarce, this might limit or impede the aims of infusing domain specialties into generative AI.

Do you see how this all adds up?

I’m sure you do, and you don’t need a GPU to do that calculation for you (pun!).

We can also include the other earlier noted prediction about the compacting of generative AI and see how that plays into this deck of cards. You might be tempted to think that maybe we have a possible band-aid or at least something that can potentially alleviate somewhat the GPU crunch. If you can push the generative AI out into the edge, you can use those devices and their computational processing to take a load off of the servers in the cloud. Great idea.

Of course, the thing is, GPUs can be included in the edge devices too, thus, you are still potentially relying upon GPUs. They are nearly inescapable these days.

The bottom line is to keep your wits about you when it comes to GPUs. You probably won’t be directly aware of the GPU issue, but you might indirectly be impacted. I suppose you could decide to create your own secret stash of GPUs, having them available to run generative AI, but that’s a different matter for a different day (you might want to stock some in your underground bunker if that’s something you are mulling over anyway).

Businesses that want to use generative AI are going to potentially be painfully aware of the GPU shortage and fierce competition. Countries are aware of this since they want to stay at the forefront of AI. Lawmakers and regulators are also tuned into the GPU malady. It is all hands on deck.

For additional details on this and related topics, see my coverage at the link here and the link here, just to mention a few.

(6) Generative AI Courts Confusion Copyrights Consternation Cajoling Congress

Generative AI has made a rather ugly, mudslinging, protracted, outsized legal and ethical quagmire.

You might have seen in the mass media the nonstop handwringing about Intellectual Property (IP) rights and generative AI. All manner of companies are suing the AI makers of generative AI. It seems that whenever a new lawsuit is launched, coverage goes wall to wall.

Here’s the deal.

One big question is whether the AI makers have overstepped the copyright laws by having pattern-matched on copyrighted content that was scanned on the Internet. To data-train generative AI, the AI makers make use of a wide swath of the Internet to find text, images, and other forms of content that can be readily used during the data-training process. Without this data, generative AI would be sorely lacking in fluency and would be of little use or interest.

Publishers, authors, artists, and all manner of content creators would assert that their copyrighted material was essentially ripped off without them getting any just compensation. If the AI makers wanted to use the content, they should have approached the copyright holders, they insist. Some copyright holders might have freely said yes, go ahead and scan their content. Others might have flatly refused. Many would probably have asked for money to compensate them for the use of their material. The money might have been a straight-out fee or maybe a negotiated commission or percentage of whatever the AI maker makes from the use of the content.

There is more.

Another concern by the copyright holders is that the scanned content at times has been so-called “memorized” by the generative AI and thus is verbatim able to be emitted by the generative AI. The gist is that the content wasn’t merely used to do pattern-matching. It was at times outright memorized or, one might suggest, precisely copied item for item. In some or many such instances, the pattern-matching opted to fully incorporate the data within the model. This seems a bridge too far, even if one allows for a willingness to do the scanning in the first place (which many do not believe such an allowance is fair).

A momentous fight is underway. Sparks are flying. Yelling and bellowing is occurring. It is a vicious wrestling match.

Into this quagmire steps an interesting turn. Allow me a moment to explain.

You might have quietly noticed a recent announcement that publisher Alex Springer opted to forge a deal with OpenAI. It seems that rather than getting mired in a long, slow, and costly legal slog in the courts as a means of ascertaining whether generative AI apps such as ChatGPT and GPT-4 of OpenAI are infringing on the intellectual property rights of publishers, it appears that Alex Springer realized that getting a piece of the action might be a more industrious approach. The deal they struck with OpenAI gives access to the content of Alex Springer so that users of ChatGPT and GPT-4 will be able to see generated summaries of presumably pertinent materials. Doing so will aid OpenAI by presenting bona fide reference material to its users, and users will be somewhat reassured that the relied-upon source is real and not a so-called AI hallucination.

What does Alex Springer get out of this arrangement, you might be asking.

The summaries will include a citation that provides a link to the otherwise subscription-required content at Alex Springer. Users clicking on the link will smoothly flow over to Alex Springer and thus become potential new subscribers. Ka-ching, the cash register at Alex Springer will be ringing as new subscriptions go through the roof, delivered seamlessly to the inviting front door of the publisher. Furthermore, imagine how extraordinarily upbeat the branding exposure will be for Alex Springer. Users will see all these peppered references to Alex Springer publications, over and over again. This will be a subtle and subliminal mental motivator that Alex Springer’s publications must be the best sources around for any kind of rock-solid factual content.

Will more of the publishers decide to switch rather than fight, in terms of cutting deals with the AI makers?

I would think so. An arduous mental calculation must be made. Will a legal battle be so delayed and costly that it would be better to cut a deal with the AI makers? The downside is that the deal would likely undermine or require an abandonment of the legal case against the AI maker. This is a tough pill to swallow because there might be vast buckets of money at the end of the lofty lawsuit rainbow. Alternatively, nothing but a lump of coal and a massive transfer of wealth from the publisher to the lawyers who valiantly fought a never-ending years-long legal dispute might also be at the end of the road.

Choices, choices, choices.

There are lots more legal questions in this murky marsh.

For example, when generative AI produces an output, a question arises as to whether the output deserves copyright protection. Suppose Joe enters a prompt into generative AI and an artwork is produced based on the prompt. Is the artwork considered copyrightable? Does Joe own the copyright or does someone else such as the AI maker? If the output looks similar to some scanned artwork that was used at the initial data training, does this “new” art infringe on the copyright of the original work?

These kinds of questions are consuming tons of time for lawyers right now.

If the AI makers prevail and can convince the courts that the scanning was considered fair use of copyrighted material, they will presumably be off the hook for potential damages of having infringed the copyrights. Some say that publishers, authors, and artists will be left high and dry with no compensation for their original and copyrighted content. Of course, another active argument is that if the AI makers do not prevail, and if they have to pay out big bucks, there is the possibility that this would undercut and possibly eviscerate the generative AI market. Almost none of the AI makers could presumably stay in business. Innovation in AI especially generative AI would be curtailed or squashed.

The counterargument is that this fanciful utter-destruction argument is a false front. Perhaps the AI makers could pay what they owe and then on a go-forward basis make sure that the copyright holders are made whole. All kinds of amenable arrangements could be worked out. The use of the classic “the sky is falling” seems calculated to keep rightful copyright holders from garnering what they are due, says those who feel the matter is being improperly portrayed.

All this legal wrangling has passionately drawn the attention of lawmakers and regulators.

Perhaps new AI-related laws are needed. But what should those AI-related laws proclaim? Stakeholders on each side of these issues would vehemently argue that their view is right, and the other side of the coin is wrong. A battle among bigwigs is taking place. The stakes are high.

Readers of my column are well aware that I have extensively discussed what Congress has been considering about AI, along with what the White House has been doing, and what is taking place at the state, county, city, and local levels. Plus, I’ve closely analyzed the European Union (EU) and its legal considerations about AI, such as the AI Act (AIA). Etc.

Existing polarization seems to consume all major issues of our day, far beyond the realm of AI alone. Trying to figure out a resolution to thorny legal and ethical issues about AI is not going to be easily attained.

I bring up all these machinations to mention that in 2024, I would predict that a huge amount of societal and legal attention is going to go toward generative AI and all its surround-sound societal pluses and minuses. This will mainly be heat and light. For example, by and large, I doubt that any of the existing lawsuits will make any substantive progress in 2024, namely it will take likely several years to see where this is going to head. The courts work on their own timetable.

With the elections coming up in 2024, the chances of reaching sufficient agreement about AI legal considerations seem rather remote. That being said, there is a notable fly in the ointment. I will address that spicy condition by doing so in my seventh prediction, shown below. Hang in there and see what the seventh prediction has to say.

For additional details on this sixth prediction about the legal and ethical aspects of generative AI, see my coverage at the link here and the link here, just to mention a few.

(7) Generative AI Deepfakes Disrupt Democracy Devilishly Dastardly Disturbingly

I saved this particular prediction for the last of the list, but I probably could have easily listed it as the first one on the list. It is that important. It is an all-consuming matter that will be at the forefront of nearly all discussions, debates, agitations, and other provocations about generative AI in 2024.

Are you mentally ready for the big reveal?

Here’s the deal.

Deepfakes is going to be the 600-pound gorilla, the infamous elephant in the room, the big cheese as it were, and will take all the air out of the room when it comes to deliberating in 2024 about the nature and control of generative AI.

This makes abundant sense because we know that the major elections are going to occur in 2024. The stakes are high in the elections. We also can easily anticipate that deepfakes are going to be completely crazy and pervasive, muddling and confounding the activities underpinning the run-up to the elections.

The deepfakes are usually devised via the use of generative AI. Ergo, all the anger and angst about deepfakes is indubitably going to land on the doorstep of generative AI and the AI makers that make and promulgate generative AI.

If we have any chance of overcoming polarization on a contemporary topic, perhaps the matter of deepfakes and generative AI is one of them. The intensity of disgust, concern, consternation, and public dismay about deepfakes is going to force lawmakers and regulators into doing something. They will find themselves in a pickle if they sit on the sidelines and merely point their fingers.

Something will need to be done.

Overall, I am saying that I think that much of the legal wrangling about generative AI will be gradually slowly winding its way forward and showcase few substantives in 2024 (things will take longer to play out), but that the deepfakes topic is going to be of such prominence that generative AI and some kind of legal stipulations involving what to do about deepfakes is bound to get traction.

The specific nature of what that legal stipulation will do or specifically encompass is a bit of a guess right now. There is a solid chance that whatever it is will be inadvertently overreaching. There is a solid chance too that whatever it is will be inadvertently underreaching. Flip a coin. The key will be that presumably action is being taken, even if the action has little to do with resolving the deepfakes conundrum.

Action will be demanded, and action will be provided.

For additional details on this and related topics, see my coverage at the link here and the link here, just to mention a few.

Conclusion

Congratulations on having made your way through the adventurous journey of my AI predictions for 2024.

As a reminder, the seven predictions consisted of these categorical areas of generative AI:

• (1) Generative AI Multi-Modal – Makes majestic mainstay manifestly moneymaker
• (2) Generative AI Compacted – Capability categorically cautiously convincingly considered
• (3) Generative AI E-Wearables – Enter evocatively everywhere eventually earnestly
• (4) Generative AI Functionalities – Fervently favor focus for fruitfulness
• (5) Generative AI GPUs – Golden gadgetry gaps gaspingly glaringly
• (6) Generative AI Courts – Confusion copyrights consternation cajoling congress
• (7) Generative AI Deepfakes – Disrupt democracy devilishly dastardly disturbingly

Take a close look again at my (slightly) poetic descriptions that say things such as multi-modal makes majestic mainstay manifestly moneymaker.

Does that mouthful make sense now?

Please say yes.

Here are the seven predictions shown in all their poetic glory:

• (1) Multi-modal makes majestic mainstay manifestly moneymaker
• (2) Compacted capability categorically cautiously convincingly considered
• (3) E-wearables enter evocatively everywhere eventually earnestly
• (4) Functionalities fervently favor focus for fruitfulness
• (5) GPU’s golden gadgetry gaps gaspingly glaringly
• (6) Courts confusion copyrights consternation cajoling congress
• (7) Deepfakes disrupt democracy devilishly dastardly disturbingly

Those are fun to read aloud. Maybe they will be memorable for you. I wanted to provide a touch of whimsy about what otherwise might seem to be a rather dry and serious set of 2024 predictions.

You can be sure of the iron-clad fact that I will be covering these topics in my Forbes column as 2024 proceeds and reveals itself. I’m sure that other topics will arise that perhaps were under the radar and didn’t seem prominent for 2024. Some surprises are bound to come out of the woodwork too. In addition, the pace and visibility of the topics I’ve discussed today will wax and wane throughout the coming year.

We live in the best of times and in the most challenging of times.

Be assured that good luck is in store for you in 2024, so please go ahead and ring the bells accordingly.