Article Summary: OpenAI CEO Sam Altman | AI for the Next Era by Samuel H. Altman

Recommendation

As artificial intelligence evolves, its ability to work alongside humans solving complex problems in fields such as medicine, finance and governance becomes increasingly critical. Examples include OpenAI’s GPT-3 language generator, image creation platform DALL-E, and code generator Copilot. OpenAI CEO Sam Altman predicts that within a few years, AI will fuel an explosion of advances in all major industries, and disrupt the global economy. Altman joined Greylock partner Reid Hoffman during the company’s recent Intelligent Future summit. Altman’s answers were prompted by questions from Hoffman and the audience.

Take-Aways

• Multiple players can create distinct businesses based on APIs created by large models.
• People starting an AI business should consider specialized training on existing models.
• Scientific research will be accelerated by APIs.
• The evolution of AI is rapid and widespread throughout human endeavors.
• Technologies like GPT-3 will impact medical and life science research.
• Users will directly interact with foundation models within five years, viewing AI as co-workers.
• Societal and creative issues will be addressed as AI influence expands.

Summary

Multiple players can create distinct businesses based on APIs created by large models.

Today, successful AI-based copywriting or educational service businesses are possible. People haven’t yet seen the “trillion-dollar take” on Google, but people are beginning to think, “How did the fundamental things change?”

“I would guess that with the quality of language models we’ll see in the coming years, there will be a serious challenge to Google for the first time for a search product.” (OpenAI CEO, Sam Altman)

Today’s chatbot interfaces work well, opening possibilities for large companies to provide medical services, education or meaningful advice. Soon, multimodal models will create more opportunities. People are working with agents that use computers for natural language communication and coding. Early versions of this include Copilot and DALL-E.

An explosion of new companies will use this type of interface or “API” on breakthrough technological platforms which people haven’t seen since mobile devices were introduced.

People starting an AI business should consider specialized training on existing models.

People will build on a handful of basic large models, but it’s unclear whether start-ups can successfully train proprietary models. What will probably happen is that many start-ups will start with an existing large model, and fine-tune it. They will create a system they can use for their own business or share, and this unique data model will improve over time, creating substantial value in this “middle layer.”

The biggest mistake in people’s thinking about AI is that these systems won’t generate new knowledge or benefits for humanity, like curing cancer or performing scientific research.

Scientific research will be accelerated by APIs.

Two critical things are happening now. One is the development of “science dedicated products” such as AlphaFold.

“If I had time to do something else, I would be so excited to go after a bio company right now. I think you can just do amazing things there.” (Sam Altman)

A second new phenomenon is “productivity tools” that indicate new research possibilities. These tools write codes, doubling an engineer or scientist’s productivity. Copilot is one example. These will alter the way scientific and technological developments occur.

But the most exciting AI trend involves fostering scientific self-improvement. AI could, for instance, automate a developer’s work, and solve tough alignment problems. The sci-fi version of that is AI self-editing code, or changing an optimization algorithm. A less scary version is what humans do when working toward scientific discoveries – they develop theories and test them. AI may be able to learn how to perform certain processes that are thus far unique to humans.

“I’m a big believer that the only real driver of human progress and economic growth over the long term is the societal structure that enables scientific progress, and then scientific progress itself.” (Sam Altman)

The “alignment problem” arises when people create a powerful system that either doesn’t do what people need it to do, or its goals conflict with human goals. In science fiction movies, for instance, AI doesn’t care much about humans. Through misuse, AI becomes a threat to humans.

“So the alignment problem is: How do we build AGI [Artificial General Intelligence] that does what is in the best interest of humanity? How do we make sure that humanity gets to determine the future of humanity?” (Sam Altman)

Today, alignment problems can be solved on a small scale. OpenAI’s largest models are fairly well aligned, but no one can predict how these problems will be solved in 100 years. Alignment research will become a crucial resource.

The evolution of AI is rapid and widespread throughout human endeavors.

Language models will go much further than many people believe. The potential for algorithmic progress is exciting. Multimodal models will move fluidly between text, images and every other modality. Right now, GPT is “stuck in the time it was trained,” but that situation is changing.

“If you just think about what…[GPT] alone is going to unlock and the applications people will be able to build with that, that would be a huge victory for all of us and just a massive step forward and a genuine technological revolution. But I think we’re likely to keep making research progress into new paradigms as well.” (Sam Altman)

New systems will help people answer questions about knowledge generation and advancing humanity. One unfortunate circumstance is that AI has become a “mega buzzword.” Lots of people are claiming to apply AI to things like solving [nuclear] fusion. But those solutions are probably not as good as those being worked on by physicists.

AI will “seep in everywhere,” even in complex systems such as financial markets. Over the next decade, the marginal costs of energy and intelligence will trend toward zero. Those heavily affect the cost of everything else, creating a seismic societal and economic shift. People who choose to spend lots of money on energy and computation will reap enormous financial returns.

Probably, the “metaverse” will eventually resemble the iPhone, a new kind of software container and interaction device. AI will in the end drive a “legitimate technological revolution.” The question will likely be: How will the metaverse fit into the evolution of AI?

Technologies like GPT-3 will impact medical and life science research.

Current AI models can’t make large impacts on medical and life science research, according to experts. These fields will be impacted when “$100 billion to $1 trillion companies” take an interest. For example, pharmaceutical companies could be transformed, but human trials would still be necessary. Synthetic bio companies could produce faster cycle times from development to product, but testing would remain a factor. Start-ups require fast cycles and low cost to be competitive in the medical field. Bio-manufacturing might be a workable field.

Users will directly interact with foundation models within five years, viewing AI as co-workers.

People will most likely discontinue prompt engineering within five years. People will interface with computers using natural language. They may use prompt engineering to ask a computer to do complicated research or provide therapeutic advice. Artists will always do better than AI with image generation.

“What will matter is the quality of ideas and the understanding of what you want.” (Sam Altman)

AGI is equivalent to a median human that you can treat as a co-worker. It could learn how to be a doctor or a competent coder, figuring out how to become adept at whatever a person needs. On the other hand, superintelligence, if it ever emerges, will happen “when it’s smarter than all of humanity put together.”

Societal and creative issues will be addressed as AI influence expands.

Society will struggle with AI’s economic impacts. Dealing with these disruptions will become paramount over the next two or three decades. People will have to figure out wealth redistribution, who has access to AGI, and what these systems can and can’t do. People shouldn’t worry too much about how their personal lives will change – those issues are solvable.

For example, OpenAI is running the largest Universal Basic Income (UBI) experiment ever undertaken. OpenAI accepts input from populations it thinks will be most affected by this new technology, in an effort to reskill people early in the cycle.

In the short term, AI is a great application for creatives. It enhances rather than replaces creativity, in large part. This trend will likely continue, at least for a few decades. Ten years ago people thought AI would first come for blue-collar factory jobs and truck drivers, then it would replace white-collar jobs, and finally, it would take over creative jobs. In fact, it’s moving in the opposite direction.

“There’s an interesting reminder in here generally about how hard predictions are, but more specifically about how we’re not always very aware, maybe even in ourselves, of what skills are hard and easy, and what uses most of our brain.” (Sam Altman)

Humans are balanced on the precipice of an AI revolution. Many people aren’t sure whether to applaud or fear AI’s emergence.

About the Author

Samuel H. Altman is an American entrepreneur, investor and programmer. He is CEO of OpenAI and the former president of Y Combinator.