Types of AI companies to come
Unless you have been living under a rock, like me you must be flipping out over all the developments happening in the AI/ML space. It feels as though the ground is shifting under our feet. There are new breakthroughs reported every week. Once again, we are heading into an uncharted future, as we did during the advent of smartphones and during the advent of the world wide web (and one which crypto failed to deliver). Those times led to the creation of generation-defining companies like Google, Meta, and Uber, to name a few. So I thought it would be an interesting exercise to try and visualize what new companies & products might come out of this next paradigm in computing.
1. B2B companies helping others build & deploy AI models
“When there is a gold rush, sell shovels” was true for the 1849 California Gold Rush, and continues to remain valid. The first class of new companies we shall see, are indeed already seeing, are those who build these proverbial AI shovels.
A successful AI lab needs 3 things — a) compute power, b) data & c) human expertise. We can expect many companies to start out offering exactly these -
Data companies — A class of companies will emerge which create & license datasets.
To create new datasets, they will employ a low-wage workforce from developing companies. In a way, these companies are ‘bootstrapping’ for AI — using human intelligence to create datasets that shall help machines become good enough to eventually replace humans.
Another rich source of proprietary training data would be created when these data companies start partnering with brands & copyright owners. Think of the company becoming an agent-custodian of an actor’s appearance and voice, or even a company & its products (eg: Ferrari’s car designs). With generative AI getting so good that it’s becoming impossible to tell actual & generated apart, the difference between ‘actual’ & ‘generated’ will dissipate. Actors will start signing off on content generated by AI in their likeness, and the rest of the world will also stop caring whether the actor has actually acted or the content is AI-generated.
AI Consultancies — Once you have the compute (hardware or cloud) and data (owned or bought), you need AI experts to create something useful. These AI experts will be like Michelin-star chefs, mixing highly-sophisticated tools & powerful ingredients in just the right recipe to create something magical, something which an average person won’t be able to replicate even if have the recipe.
What will an average business need AI consultancy for? The usual — demand prediction, inventory management, pricing, recommendations & targetted advertising, anomaly detection, fraud detection, risk analysis, and customer support.
Infrastructure companies (IaaS, PaaS for ML) — This is a capital-intensive business — whether it be providing Cloud services or providing customized ML chipsets for on-premise computing. The legacy players — TSMC, Nvidia, Qualcomm, ARM, Intel for chipsets & Azure, AWS, and Google for cloud — will continue to dominate each market. Each company will innovate at a furious pace, fearful of being left behind.
ML Models marketplace — Developers would prefer to use off-the-shelf models for generic tasks. AI outfits would emerge which specialize in solving a particular narrow problem and then build a very specialized dataset to train their model. They will start serving this model for others to use. They will offer different degrees of customization to enable transfer learning to new tasks. The pricing power these companies command will come from how proprietary the data they have. In the absence of proprietary data, their pricing would devolve to a cost-plus model, wherein they only charge a small premium over the compute needed to serve every API call.
2. A new paradigm for creativity
No other place had the ground shaken more than for the creatives. The developments from systems like ChatGPT, StableDiffusion, Face2Face, are so remarkable I personally would have considered them science fiction just a couple of years back.
What does this mean for people in the creative arts?
Supercharged Creative tools — All tools of creative expression will become insanely powerful. All “creation” activities will change to “editing” activities. We do this already with autocorrect & email suggestions. AI is already suggesting us how to complete our half-finished sentences. A simple extrapolation can help us imagine the final form of this.
Imagine you are writing your first novel and have enabled the “AI Helper”. Based on an initial prompt, the AI already has finished a novel for you. You start reading AI’s creation until you find something which doesn’t match your expectation for the story. You make a change there and the AI immediately updates the entire rest of the novel to keep it consistent. You continue repeating this process over and over until you have a draft you are happy with. So instead of writing the novel, you just kept editing the AI’s creation.
What is the long-term impact of this on human creativity? That’s a difficult question to answer, but the genie has come out of the bottle & will not go back.
Copyright Management Companies — Every creator, whether it be a writer, artist, actor, musician, vlogger will start copyrighting their ‘essence’. The definition of ‘essence’ will evolve with time, and copyright laws will need to evolve in tandem. For people at the top of their game, this would become another big source of revenue as they can sit at home and let their ‘essence’ be used in advertisements, games, movies & everything else.
Creators vs Automation — The fight for survival is going to become even more intense for creators as entry barriers are lowered. Any Tom, Dick & Harry like me would be able to enter any creative-domain.
We will see a plethora of simulated celebrities on film, games & social media. Conceptually these are nothing different from animated characters, except they look life-like.
Not only would a creative need to contend with this AI-enabled competition, but they would also need to compete with the superstars more directly. An A-list celebrity like Tom Cruise may not have bothered about a local advertisement since it’s not worth their time, but what if someone could license this actor’s “essence” for a fee? The celebrity can just sit at home while their essence gets used in every local kirana store across the country. The celebrity may even retire or pass away, but their “essence” will not. Capitalism will not let a brand die that still has value.
As the competition intensifies, recall value becomes more important. Hence, marketing departments & power brokers who control the entry gates will become even more important.
3. Impact on Journalism
With DeepFake & style-transfer technologies, AI-created content would be indistinguishable from reality. There is a real worry of fake news running rampant as it becomes impossible to tell what is real.
Maybe I am being an optimist, but upon deeper reasoning, I feel the problem of fake news may actually reduce as people will become much more aware of it. They will have to learn to go to trusted media houses to get genuine information. An analogy here is how Covid actually reduced the total deaths in the USA since people went out less & there were fewer cases of other flu & car crashes, etc. Similar to how covid forced people to be more hygienic, FakeNews might educate people to get their information from only validated news outlets.
However, given the above, the power of big media houses would increase. The indie outlets would find it really hard to retain trust. They would find it increasingly difficult to break controversial news or report sting operations, as any such attempt would get drowned down in the torrent of FakeNews. And the big media houses are more susceptible to capitalist cronyism & government diktat. All in all, this can lead to a situation where authoritarian governments have stronger state control over the media and become more totalitarian.
4. Impact on other industries
Recordable expertise
If a profession’s input & output can be captured in a fairly neat manner, then this job would be at risk of automation. New products and companies will emerge offering automation for every such profession.
Unfortunately, this puts a lot of freelancers at risk.
For company people, there is a moat in the form of a company-specific context. People spend many years building this context, becoming more and more valuable for the company. This context has the humanities part — culture, relationships, management style, etc — and the hard expertise part — internal codebase, data structures, etc. The humanities part feels almost impossible for today’s AI systems to subsume & would be a final bastion of human-labour. But leaving aside the humanities part, automation of the hard-expertises still puts many jobs at risk. Think of how fast a new employee start becoming productive within the company. Now imagine that a computer can process information a million times as fast, doesn’t sleep or rest, and once trained, can be replicated many times over. To be honest, I am a bit afraid for my own role & my friends who are data analysts & programmers. I don’t expect these jobs to disappear entirely, but the tools for them will become so much more powerful, that you would not need a high degree of expertise to operate them. Accordingly, the pay in these jobs would reduce and there would be fewer individuals needed in these professions.
Who is safe?
Even with a lot of research going over it, general-purpose intelligence is quite a bit further away. This means that AI still needs tasks to be presented to it in a relatively clean manner and the input/output needs to be fairly well-templatized.
Current AI systems cannot handle even simple tasks in an unstructured environment. A travel agent who looks at multiple websites and route options to plan your trip is not getting automated anytime soon. This is because while the problem statement is relatively simple & well-defined, the websites of different airlines, hotel & local amenities are not standardised enough for one AI to look up information from all of them. Only once AI learns to navigate through disparate website interfaces naturally like a human can our travel agent be automated.
Mechanical manipulation also looks far from reality at the moment for multiple reasons. Training in the real world with real physics is slow (though Google is trying it). Training in a physics simulation is fast but not accurate enough. So if your profession requires mechanical manipulation in an unstructured environment, then your job should be safe for the time being. This covers baristas, chefs, warehouse workers, etc. Especially the barista, try as we might, the art of developing a good coffee machine still eludes us.
Jobs involving ‘deep’ connection with other humans are also not going away. But jobs with only a transactional, evanescent interaction will. The 5-minute one-off support-center phone call will most definitely be automated away.
5. New technologies
AI is an enabler and I am really excited to see its applications in solving some of the important problems in the world, beyond targetted-advertising and dopamine-hacking recommendation engines.
Search & Retrieval — Our relationship with information is changing, yet again. Until pre-enlightenment, we had to work with the information we had in our heads. Then the printing-press came and our information frontier expanded to all libraries we have access to. Then www & google came about, and we got all the world’s information at our fingertips. With social media & instant messaging, access to information started meaning who is in your network whose expertise you can tap into.
Now, with large-language models we are at a new frontier. An expert in every field in available at our fingertips. There are definitely some teething problems — like the AI confidently spewing fluent bullshit when it doesn’t know the answer — but these will surely get solved with time.
How will this change our education system? Will this encourage or discourage people from becoming experts?
Search within Audio & video — It is going to be really exciting to see how Search and Generative-AI tackle audio & video, which it surely will. The impact may be even more significant here since the baselines are so low. For text, we still had a decent search engine, even Ctrl-F works decently for many tasks. But for audio & video? The best we can do is search within the annotated text. For audio, I can’t skip semantically, cannot ask the player to skip ads, or move the audio to the interviewer’s next question. For video, I can’t describe the scene and ask the player to take me there (Eg; Take me to the Avengers scene where Hulk punches Loki). With the pace of AI as it is, all these capabilities likely already exist in research labs and may land in our hands in another couple of years.
Summarisation — So much of the mundane work in corporate will get eliminated if there is a simple summarisation tool that can automatically send out meeting notes and generate a gist of verbose strategy documents. Personally, this might be the advance I am most looking forward to.
NLP + Machine Translation + Airpods : Each one of us shall have a combination of the universal translator & Jarvis in our ears very soon. The building blocks are already there.
Healthcare — AI in Healthcare is one to look out for & it deserves a post of its own. Apple felt the need to come out with a 60-page dossier on what it is doing in Health. The opportunity here is enormous, particularly in the diagnosis & early detection of diseases. At an individual level, this would mean your data (biometrics, activity data, search & mobile usage) gets passively analysed to proactively warn you of any worrying trends. At a population level, constant tracking of anonymised citizen data can be used to detect spread of new infections and diseases.
Gaming — It’s exciting to anticipate what AI will do for games. In games, not only can it generate new scenes & rich open world environments on the fly, it can also populate those worlds with unique creatures imbued with a realistic personality.
Note that the list here is not exhaustive at all, for I am not an expert. I am a fly on the wall, a dilettante. There is amazing work happening in Robotics for example, which I don’t feel confident writing about so it does not find any mention above. If you are particularly passionate about an area, do leave a comment as I would love to learn about it.
