With continuing advances in machine learning, it makes less and less sense to compare AI to the human mind.
Speech and language are central to human intelligence, communication, and cognitive processes. Understanding natural language is often viewed as the greatest AI challenge — one that, if solved, could take machines much closer to human intelligence.
In 2019, Microsoft and Alibaba announced that they had built enhancements to a Google technology that beat humans in a natural language processing (NLP) task called reading comprehension. This news was somewhat obscure, but I considered this a major breakthrough because I remembered what had happened four years earlier.
In 2015, researchers from Microsoft and Google developed systems based on Geoff Hinton’s and Yann LeCun’s inventions that beat humans in image recognition. I predicted at the time that computer vision applications would blossom, and my firm made investments in about a dozen companies building computer-vision applications or products. Today, these products are being deployed in retail, manufacturing, logistics, health care, and transportation. Those investments are now worth over $20 billion.
So in 2019, when I saw the same eclipse of human capabilities in NLP, I anticipated that NLP algorithms would give rise to incredibly accurate speech recognition and machine translation, that will one day power a “universal translator” as depicted in Star Trek. NLP will also enable brand-new applications, such as a precise question-answering search engine (Larry Page’s grand vision for Google) and targeted content synthesis (making today’s targeted advertising child’s play). These could be used in financial, health care, marketing, and consumer applications. Since then, we’ve been busy investing in NLP companies. I believe we may see a greater impact from NLP than computer vision.
What is the nature of this NLP breakthrough? It’s a technology called self-supervised learning. Prior NLP algorithms required gathering data and painstaking tuning for each domain (like Amazon Alexa, or a customer service chatbot for a bank), which is costly and error-prone. But self-supervised training works on essentially all the data in the world, creating a giant model that may have up to several trillion parameters.
This giant model is trained without human supervision — an AI “self-trains” by figuring out the structure of the language all by itself. Then, when you have some data for a particular domain, you can fine-tune the giant model to that domain and use it for things like machine translation, question answering, and natural dialog. The fine-tuning will selectively take parts of the giant model, and it requires very little adjustment. This is somewhat akin to how humans first learn a language and then, on that basis, learn specific knowledge or courses.
Since the 2019 breakthrough, we have seen giant NLP models increase rapidly in size (about 10 times per year), with corresponding performance improvements. We have also seen amazing demonstrations — such as GPT-3, which could write in anybody’s style (such as Dr. Seuss-style), or Google Lambda, which converses naturally in human speech, or a Chinese startup called Langboat that generates marketing collateral differently for each person.
Are we about to crack the natural language problem? Skeptics say these algorithms are merely memorizing the whole world’s data, and are recalling subsets in a clever way, but have no understanding and are not truly intelligent. Central to human intelligence are the abilities to reason, plan, and be creative.
One critique of deep-learning-based systems runs like this: “They will never have a sense of humor. They will never be able to appreciate art, or beauty, or love. They will never feel lonely. They will never have empathy for other people, for animals, or the environment. They will never enjoy music or fall in love, or cry at the drop of a hat.” Makes sense, right? As it turns out, the quotation above was written by GPT-3. Does the technology’s ability to make such an accurate critique contradict the critique itself?
Many believe true intelligence will require a greater understanding of the human cognitive process. Others advocate “neuromorphic computing,” which is building circuitry that more closely resembles the human brain, along with a new way of programming. Still others call for elements of “classical” AI (that is, rule-based expert systems) combined with deep learning in hybrid systems.
I believe it’s indisputable that computers simply “think” differently than our brains do. The best way to increase computer intelligence is to develop general computational methods (like deep learning and self-supervised learning) that scale with more processing power and more data. As we add 10 times more data every year to train this AI, there is no doubt that it will be able to do many things we humans cannot do.
Will deep learning eventually become “artificial general intelligence” (AGI), matching human intelligence in every way? I don’t believe it will happen in the next 20 years. There are many challenges that we have not made much progress on — or even understood — such as how to model creativity, strategic thinking, reasoning, counterfactual thinking, emotions, and consciousness.
I would suggest that we stop using AGI as the ultimate test of AI. Soon deep learning and its extensions will beat humans on an ever larger number of tasks, but there will still be many tasks that humans can handle much better than deep learning. I consider the obsession with AGI to be a narcissistic human tendency to view ourselves as the gold standard.
This article first appeared on WIRED UK website on Dec 16, 2021 for “The WIRED World in 2022” special edition