Happy New Year's -- 2024

2024 is guaranteed to be another year of exponential change with AI continuing to be the primary source of acceleration and humanity's struggle with adapting to change being the primary brake. I started this newsletter in September of 2022 for two reasons -- first, I was a year+ into working with OpenAI technologies including GPT-3 and DALL-E and was impressed with the speed at which these tools were maturing. Second I was surprised that a majority of my friends and colleagues had no idea what was happening in this new domain of generative pre-trained transformers and I hoped to bring what I was learning to that larger audience in order to get them as excited about the potential for this technology as I was.

Of course November 30, 2022 OpenAI released ChatGPT, which immediately made this technology both accessible and comprehensible to basically every Internet user on the planet and my second goal of encouraging interest in GPTs was eclipsed by the sudden and complete media coverage of the topic. And so the focus for me has developed into coverage of the still evolving technology and pragmatic advice on using it. At the end of 2022 I wrote "Goodbye 2022, hello 2023" reflecting on the "3.5" major technology breakthroughs of 2023 and predicting that in exponential fashion we should expect more than one breakthrough a month during 2023. I think we can all agree that whatever the measure of "major breakthrough" might be, that we exceeded that goal of 12.

So I have taken the idea of 12 and reframed it as themes -- 12 developments in 2023 that we should expect to continue into 2024 as this exponential pace continues. So here are the things I think we should be reflecting on from 2023 and watching continuing development of in 2024:

Multimodal

Before 2023, pre-trained transformers were already being applied to different types of information (words, code, sound, music, images...) but the important breakthrough was the release of models which could handle multiple types in one system. Expect this to continue to pay dividends in 2024 as these models become more capable and useful for a broader set of tasks.

Competitors to OpenAI/Microsoft: Google, Apple, Meta, IBM?

While OpenAI (and by extension their partner Microsoft) continue to have a surprisingly large lead over others, 2023 was full of announcements of new entrants into the pre-trained transformer marketplace. Basically all of the major technology companies introduced their own OpenAI competitive platform. The challenge for each in 2024 will be to catch up with and surpass (at least in some critical characteristic) a likely continuing rapid evolution of OpenAI's offerings.

Open source models

While Meta brought attention to the idea that an open source model could be competitive with commercial models, 2023 saw hundreds (thousands?) of efforts launched to produce open transformer models and training data, making the technology accessible beyond the multi-billion dollar efforts of the larger companies. We should expect this to continue in 2024 with positive and negative consequences -- models will be developed for niche use cases that support specific research initiatives and community needs, but also for use by bad actors or unsavory activities.

New advances in particular modalities: Image, Music, Video, Dance?

Despite the importance of multimodal models, one area where saw more specialized models improving was in their focus on specific information types. 2023 brought impressive improvements in image generation, music creation, the first steps toward video production, and even new types of information like choreography for dance. Expect that this will continue in 2024 with images, music, and video all reaching a stage where the AI output is indistinguishable from "real" (human created) content.

Healthcare

Despite concerns about the safety of using AI to diagnose illness, we still have millions of people in the US (and billions around the world) with inadequate access to healthcare, so it wasn't surprising to see tremendous advances in 2023 in using AI for a wide range of health applications. From cancer diagnosis to routine checkups, demonstrations of the careful application of pre-trained transformers are showing that AI can be as good (or better) than humans at providing health outcomes. Major players in every part of healthcare are focused on the application of this technology to reduce costs and improve outcomes and 2024 will see practical real world deployment.

Teaching and learning

The early disruption for education accelerated in 2023 with healthy debates about the role of this technology in the classroom. Stanford has a longer exploration of the topic that is worth reading: https://hai.stanford.edu/news/ai-will-transform-teaching-and-learning-lets-get-it-right Suffice to say that 2024 will be a year when we have to reevaluate all of the beliefs we have about teaching and learning. Organizations like Kahn Academy with their Khanmigo AI tutor are showing a promising direction. If the goal of education is at least in part to prepare young people for participation in the workforce then teaching them how to properly use these tools should be a priority for teachers.

AI for physical robotics

Multiple breakthroughs in 2023 demonstrated that combining generative pre-trained transformers (especially multi-modal computer vision and text models) with physical robotic systems offered exciting new capabilities for these systems to navigate human environments, interact safely with people, and take on more complex tasks with less programming. In 2024 we will see more widespread use of humanoid robots (a dozen companies are working on promising avenues) initially in industrial environments, but quickly in everyday settings such as hospitality and entertainment.

Social Media influencers

All of these advances will also have questionable applications - starting with the benign: promotion of products and services through "influencers" who are entirely AI generated. While the use of invented personalities by marketers dates back to the beginning of advertising, companies like 1337 have taken this to another level in 2023 by constructing a huge number of entirely artificial personalities which, thanks to image, video, and sound generation capabilities, can seem to have complex human-like lives and interactions. We will see this scale up in 2024 to an emerging set of AI celebrities in different categories - music, fashion, sports, etc.

Manipulation

Perhaps more worrisome than having your favorite music coming from an entity which you know is entirely technology based will be NOT knowing that you are interacting with a machine. In 2023 there has already been an explosion of "deepfake" images, videos, and sounds with the intention of manipulation -- ruining a reputation, to separate people from their money, vote or act in a certain way... Going into the 2024 Presidential election year in the US we can expect that increasingly sophisticated digital manipulation will become common - don't believe anything might become the best defense.

AI Doomer / Anti-Doomer

Over the course of 2023 we saw an increasing number of voices warning about both the near term and further out risks of these technologies, as well as a backlash of equally compelling voices that point out the societal benefits for continuing research, development, and application of these technologies. Even in this short list a few of those arguments are clear -- the danger of manipulation vs. the benefit of more accessible healthcare as two examples. I expect that these arguments will continue to be refined and expanded in 2024 especially as we have concrete examples of both the dangers and benefits.

Lawsuits

A joke making the rounds in silicon valley is that the only category of hiring in 2023 bigger than engineering at OpenAI is in hiring lawyers. Legitimate questions are being asked about the training data used in these models and whether human creators of content used to train machine intelligence should expect compensation of some kind. The New York Times is the most recent entrant in the copyright lawsuit game (really good summary here on Techdirt). These lawsuits will continue to move through the legal system in 2024 regardless of their merit and may also result in changes in the laws that govern both copyright and how these systems are used.

Government Regulation

Which brings us to the continuing evolution of regulation which will prove to be a difficult topic at every level of government since as a society while we want protections in place for certain things that we value, we also want the benefits that these technologies promise to bring in the future. The White House Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial lntelligence is a great example of the limitations of government regulation. It tries to walk the line between protecting citizens from the misuses of AI while avoiding interference with this important driver of economic improvement. As a result it becomes more of a plea to the industry to "do the right thing" than true regulation. Even the recent heralded AI law passed by the European Union does nothing in 2024 other than provide a warning to AI companies that stronger oversight will be in place by 2025... And by then the competitive reality will dictate more than politicians how much interference with AI technologies is practical.

What do you think will be the most important themes for AI in 2024? How will it change your business and your life? Will you be ready and adaptable? These twelve themes are certainly not the only things to watch, with plenty of surprises ahead as well. But hopefully it provides at least an initial set of topics for you to think about and be ready for as we start another amazing exponential growth period -- Happy New Year!