Endings naturally precipitate reflections on the events that came before them. At the end of the year, it’s normal to look for milestones– the significant moments that allow us to chart our progress.
Only the long view of history can reveal whether our attributions are correct; however, history might bear out that 2022 was a red letter year for artificial intelligence.
The year brought some major algorithmic breakthroughs. In the early part of the year, Meta’s data2vec– a high-performance self-supervised algorithm for multiple modalities– outperformed the best-single purpose algorithms, arguably a paradigm shift in self-supervised learning. Around the same time, Google’s DeepCTRL created a new method for training deep neural networks, allowing developers to implement and adjust rules without retraining their models. And in October, Deepmind’s AlphaTensor became the first AI system capable of discovering new, efficient, and provably correct algorithms for matrix multiplication.
During the latter half of the year, the release of Stable Diffusion shook up the art world, bringing generative AI into mainstream conversations, and the year ended on a bang with the release of ChatGPT-3, whose witty, eerie, and seemingly intelligent interactions have both enthralled and spooked the public. The platform gained 1 million users in just five days.
The already fast-moving pace of technological change witnessed over the past 12 months will only accelerate in the coming years. Considering the global events and technological milestones of 2022, here are some AI trends to watch out for in 2023.
Trend 1: Progress in generative AI will erupt, disrupting more creative industries.
The transformational leap forward in the general availability and ubiquity of generative AI that occurred in 2022 is only the beginning. The models will continue to improve, and as they do, they’ll increasingly disrupt creative industries.
In 2022, generative AI affected the fields of visual art and graphic design the most. However, companies are increasingly focused on developing models that can produce photorealistic outputs, including facial and body movements. Companies like Synthesia have already made strides in this area. As models improve, we’ll see text-driven video creation.
Technology is also poised to disrupt the music industry yet again. The same tools that are being used to generate visual art from text prompts will be expanded and applied to audio. There are already models that use text to generate music and realistic human voices, and once they start performing well enough that the public takes notice, progress in the field of generative audio will accelerate rapidly. Soon, fans of Taylor Swift and Eminem will be able to ask AI to generate a new track that combines the best features of both musicians.
When it comes to writing-based industries, ChatGPT recently showed us how well AI can mimic style and tone–whether it’s crafting press briefings, writing limericks, or constructing “how to” guides in Biblical verse.
The big takeaway? Grammarly was just the beginning of AI-assisted writing. As machine learning models improve, their mimicry won’t be used only for entertainment purposes: people will increasingly have access to the writing know-how needed to communicate clearly — and creatively.
Until recently, different areas of machine learning have been siloed, with individual models that generate images, text, audio, video, etc. However, the industry is moving towards a place where a single model will be able to generate all of those different kinds of media. Before long the combination of these technologies will lead to models that can produce AI-generated music videos, films, and more in which the machine generates the script, vocals, scores, and visuals.
In a few years, users could have the power of an entire Hollywood studio on their computer.
AI won’t replace creative industries in the next 12 months. However, visual artists are already feeling the impacts of 2022’s advancements in generative AI, and creators in other artistic fields should prepare to experience a similar disruption in 2023.
Trend 2: Society will struggle to keep up with the pace of technological change
Generative AI has captured the public’s imagination. What we learned from 2022 is that even in tough economic times, VC funding follows where the public imagination goes. (Within two months of publicly releasing Stable Diffusion, Stability AI raised $100 million at a valuation of $1B.)
As a result, we’ll see a spike in business activity and startups created around generative AI models. Entrepreneurs will spot new use cases, ML engineers will continue to innovate, and the already breakneck speed of AI advancement will accelerate.
Society will struggle to keep up. 2022 has already left some communities frantically trying to keep up with the new reality brought on by generative AI.
ChatGPT-3 isn’t necessarily the last nail in the coffin of the college essay, but academia must now struggle to decide how to monitor students using AI to produce content capable of earning them at least Cs. For the humanities disciplines, which award degrees based on composition assessments, the future is suddenly full of uncharted dilemmas.
Having discovered that models trained on their artwork without permission and users can now use generative AI to plagiarize their artistic styles, artists are trying to figure out how to protect their compensation streams. More ethical questions are also arising as art awards are given to AI-generated work and institutions promote the output of these technologies rather than commissioning and showcasing artist’s work. (The San Francisco Ballet highlighting on social media that Midjourney generated beautiful Nutcracker posters was particularly grating for many in the arts community).
At the same time, the advancement into AI-generated photorealistic images has led to an increased concern about the ability to produce deep-fakes of individuals in compromising situations.
Copyright law, obscenity laws, and fair use doctrines weren’t written with AI in mind. At the moment, the question of how these laws interact with artificial intelligence is still under debate.
Unprecedented questions are arising around ownership. For instance, if AI can easily replicate a person’s face, then who owns the likeness of a face? Should an artist receive royalties if a model trains on their artwork? If a bot produces literature or poetry, does the user have a right to claim authorship?
Resolving these issues requires moving slowly through due process and regulatory systems. All the while, these technologies will continue to advance, outpacing regulatory frameworks and guidelines. Given the tension between the pace of regulation and the pace of AI advancement, policymakers will have to walk a careful line, crafting policy that protects consumers against agents misusing the technology while not stifling technological progress that can ultimately benefit millions of people.
In the meantime, companies will likely attempt to regulate themselves, updating their AI products so that they cause as little harm as possible while still pushing the technology forward.
Trend 3: AI systems will move closer towards self-improvement
The past year has also seen the development of models that are improving the way that elementary mathematical operations, such as matrix multiplication and factorization, are computed.
In 2023, continued progress will likely lead to AI systems that can self-improve other AI systems. Matrix multiplications make up the bulk of the “compute work” involved in machine learning, so if a model can develop more efficient approaches to matrix multiplication, then it has the potential to exponentiate the current rates of progress in ML models.
The task of building more efficient and powerful models will no longer rely on human ingenuity alone: there will be models that seek out and implement these improvements themselves. Improvements in model performance at this scale could be the innovation that leads to AI discovered medicine or AI generated Hollywood movies.
As AI systems begin to solve these problems on their own, ML engineers will have their time freed up to focus on seeking out new use cases and applications for machine learning.
Trend 4: The AI arms race will accelerate
The backdrop against which 2022’s AI advancements occurred was one of a rapidly shifting geopolitical landscape. Geopolitical tensions have continued to escalate throughout the past year, and they’ll remain elevated in the year to come.
As a result, we will see an extension of the AI arms race as well as changes in how governments view, use, and regulate AI.
Russia’s invasion of Ukraine has likely accelerated governments’ investment into AI projects related to national security, particularly in relation to satellite and aerial imaging. In the coming year, governments are going to become a lot savvier about how they collect and use the data at their disposal in support of their national security agenda. Nations will be watching how their allies and their adversaries use data to develop AI, and they will move quickly to mimic those strategies.
These pressing national security needs mean that governments will pour investment into artificial intelligence, viewing AI as strategic funding for their defense systems. They’ll also work to attract the best and brightest minds from academia and industry to work on these AI projects.
As government investment in AI grows, so too will the need to create policy that confronts rapid technological advancements. As a result, protectionist policies will likely increase as will governments’ interest in regulating AI more generally.
While 2023 is set to be a year full of exciting development in artificial intelligence, the pace of progress will create challenges for individuals, communities, and nation states. AI experts, policymakers, entrepreneurs, community advocates, and other key stakeholders should be working together now to best figure out how to navigate the multifaceted changes that will accompany the AI revolution.
