“AI will disrupt every industry”

Richard Socher, people tend to think they know what artificial intelligence (AI) stands for, but it might be helpful to get your definition. What exactly is AI?
We define it as a field of computer science that tries to at first mimic human intelligence, but eventually also expands beyond our capabilities. There are different kinds of intelligence: visual intelligence, which is tied into computer vision, so that computers understand images, videos and the visual world. Then there is motor intelligence. That’s not as unique in the animal kingdom, but it's still non-trivial to replicate, and is connected to the field of robotics. Natural language understanding is, to my mind, the most interesting manifestation of human intelligence. Trying to understand and also generate language is the one intelligence that separates us the most from other animals and also connects deeply to culture, history, knowledge, thought and everything else. Financially speaking, it’s also the most impactful field of artificial intelligence right now. Search engines are based on natural language, so are advertising and translation.

Has such smart software already bested its human creators?
We've expanded artificial intelligence and are doing things that no human could do, for instance, reading a sequence of proteins, and being able to create a brand-new protein. Humans have not evolved to read sequences of amino acids and understand the language of proteins, but artificial intelligence can read billions of documents or billions of protein sequences. In biomedical applications, artificial intelligence is already much superior to human intelligence. The same is true for a lot of other things, like business processes. When artificial intelligence can read millions of documents, it can understand how often an engine fails and perform predictive maintenance.

There’s another flavor of alternative intelligence that is getting a lot of attention right now: generative AI. How is that technology different from what you just described?
The difference between generative artificial intelligence and the previous wave of intelligence was that it was much better at understanding inputs that we can use. Now, artificial intelligence has also gotten really good at creating these inputs. You can use generative AI to have it write code, essays and articles for you. And you can use artificial intelligence to generate images. Music will be next. Eventually, artificial intelligence will be able to generate complex movies. Generative AI has made a ton of progress recently, hence the huge excitement.

Google and Microsoft are eager to spend billions to incorporate intelligent chat into search. You also founded a search engine called you.com that now has a chat function. How will search change?
In the next six months or so, search will look nothing like it did for the last 20 years. It will be much more useful, interactive and powerful. Not just a list of blue links, but much more chat-based, where you have a simple prompt to ask questions.

Other than chatbots and searches, what are the most promising applications for artificial intelligence that are already in use or on the horizon?
Artificial intelligence will disrupt every industry that can collect any kind of data and has repetitive processes. In agriculture, for example, you can have a tractor drive through a field and identify what's a weed, what's lettuce, and then only pick the weeds out or spray them with a tiny amount of herbicide, and use less water. Then there’s healthcare. Automation and artificial intelligence can make radiologists more efficient by doing simple things like reading an X-ray. And finally, office work. Every time you have to write an e-mail, marketing materials or need a picture, artificial intelligence can help you. If you want to creatively brainstorm new ideas or questions, you can ask an artificial intelligence expert. You can name any industry and I can come up with examples of how artificial intelligence could improve that industry.

Are there really no professions where alternative intelligence won't play a role for the foreseeable future?
It's mostly industries that don't generate enough revenue and don't have enough repeatable processes. Individualized house cleaning, for instance. It’s a very tricky thing to collect a lot of data since every house is different. Every household sorts their laundry slightly differently and has different nooks and crannies. No one will collect 100 000 examples of how to clean my house, and it will therefore be harder to have any artificial intelligence built for housecleaning. Also, robotics has been struggling with fine motor skills. That’s why robotics and artificial intelligence have not made as much progress in surgery as you would think. There are also a lot of jobs where you want human interaction – empathy and sympathy from someone. Those will continue to thrive even more.

Most experts agree that the combination of artificial intelligence, automation and robotics will do away with a lot of blue- and white-collar jobs. What will those people do?
In the short term, we need to come up with solutions for lifelong learning. The idea that you learn a skill in your 20s and then over the next 30 to 50 years just execute on that skill and don't learn that many new things will not be viable. Humans possess a sort of swarm intelligence for coming up with new ideas, so I'm very optimistic in the long term that people will find new, interesting things to do. We will see more and more jobs evolve as it becomes easier for more people to create new apps, projects and programs.

Technology advances very quickly and often faster than policymakers or regulators can respond. What are your thoughts on the urgency of putting in guardrails for AI development?
I think it's hard to regulate the fundamentals and the research. But, of course, for certain applications, it makes a ton of sense to regulate alternative intelligence, for example, in the medical space. It’s incredibly hard building something brand new, and the fewer constraints you have, the more likely you'll be able to actually make it happen. Once it works, you should think about and regulate the specific applications, whether it’s transportation, health care, banking, insurance or especially warfare.

AI systems have been shown to make mistakes based on real-world, human bias. Can software be ethical, or are we talking about correcting the mistakes of its programmers?
In many ways, artificial intelligence is only as good as the people, the systems and the data that influence it. Alternative intelligence is a dual-use technology. You can teach the same algorithm to classify images of cancer or brain scans, or you can train it to know whether to shoot using drone images of tanks. That's where the regulations and ethics come in. Whenever artificial intelligence touches people's lives, it’s important to think about how the AI’s output affects people. Ethical artificial intelligence captures a very broad range of these applications, and examines whether artificial intelligence amplifies human biases. For instance, is a system being trained with historical data that has potentially racist or sexist patterns, and then automating decisions today based on that biased data? We have seen that happen in banking and the judicial system.
 

About Richard Socher
German-born computer scientist Richard Socher (39) is considered one of the rising stars of alternative intelligence. After receiving a PhD from Stanford University and winning an award for his dissertation on deep learning and natural language processing, Socher decided not to pursue a career in academia, and instead set his sights on commercializing his insights. His AI start-up MetaMind was acquired by Salesforce in 2016, where Socher worked as Chief Scientist. After leaving the enterprise software company in August 2020, he launched a new search engine called you.com, and founded AIX Ventures to invest in promising AI companies.