Machines are getting smarter year by year, but artificial intelligence hasn’t yet responded to the hype caused by some of the world’s largest tech companies.
AI excels at certain narrow tasks such as playing chess, but struggles to do multiple things well. For example, a 7-year-old child has much broader intelligence than all AI systems today.
Edward Grafenstett, a Meta AI researcher who previously worked at Facebook AI Research, told CNBC that AI algorithms are useful for solving individual tasks or problems involving small degree of variability.
But there is a lot of potential for change in the real world, a dynamic that training algorithms can barely grasp and lead to fragile intelligence, he added.
AI researchers are beginning to show that there are ways to efficiently adapt AI training methods to changing environments and tasks, resulting in more robust agents, Grefenstette said. He believes that this year there are more industrial and scientific applications of such methods that will result in a noticeable leaps.
AI has a long way to go before something like intelligence can be achieved on a human scale, but Google, Facebook (Meta), and Amazon are spending billions of dollars to hire talented AI researchers who has the has the potential to improve everything. From search engines and voice assistants to the so-called “Metaverse” aspect.
Beth Singler, an anthropologist studying AI and robots at the University of Cambridge, told CNBC more about the effectiveness and reality of AI in what is now known as Metaverse in 2022. Funds are invested in the region and the public begins to recognize “metaverse” as a term and concept.
Singler also warned in 2022 that the impact of the Metaverse on people’s “identities, communities and rights” could be “too little controversial.”
The cycle from lab discovery to practicality can take years,” he said, adding that there is still a long way to go in the field of deep learning. Deep learning is an area of AI that seeks to mimic the activity of layers of neurons in the brain to learn to recognize complex patterns of data.
Marcus believes that the most important challenge AI is currently facing is “finding a good way to combine all of the world’s vast knowledge of science and technology” and deep learning. At this point, deep learning can’t use all of that knowledge, instead sticks to learning everything from scratch, he said.
I expect progress on this issue this year, Markus added. This will ultimately shift towards what I call a hybrid system, but it will take several years before we see big dividends. What we’re likely to see this year or next is the first medicines where AI has played a major role in the discovery process.
DeepMind’s next steps
One of the biggest AI breakthroughs in recent years came from Alphabet’s London-based laboratory, DeepMind.
The company has successfully developed artificial intelligence software that can accurately predict the structure in which proteins will fold in days, solving a 50-year “difficult challenge” that could pave the way for a better understanding of disease and drug discovery.
Neil Lawrence, a machine learning professor at the University of Cambridge, told CNBC that he expects DeepMind to tackle more important scientific problems in 2022.
The language model (an AI system that can generate compelling texts, talk to people, and answer questions) will also be improved in 2022.
The best-recognised language version is OpenAI`s GPT-3 however DeepMind stated in December that its new “RETRO” language version can beat others 25 times its size.
Machine learning scientist Catherine Breslin, who previously worked on Amazon Alexa, believes Big Tech will move towards and larger language model next year.
Breslin, who now heads AI consulting firm Kingfisher Labs, told CNBC that they tend to move towards a model that combines vision, speech and language capability rather than treating them as separate tasks.
Nathan Benaich, a venture capitalist at Air Street Capital and co-author of the annual State of AI report, told CNBC that next-generation companies will use language models to predict the most efficient RNA (ribonucleic acid) sequences.
Last year we saw the impact of RNA technology, many of which are based on this technology, with new coronavirus vaccines ending nationwide lockdowns, he said. I believe this year will see a new harvest for AIfirst RNA Therapy Company. By using language models to predict the RNA sequences that will be most effective in treating a disease of interest, this new company can significantly reduce the time it takes to discover new drugs and vaccines.
Ethical concerns
Many advances may be imminent, but there are major concerns about AI ethics. This can be very discriminatory and biased when trained on a particular dataset. AI systems are also used to power autonomous weapons and generate fake pornography.
Verena Rieser, a professor of conversational AI at Heriot-Watt University in Edinburgh, told CNBC that 2022 will be more focused on the ethical issues surrounding AI.
I don’t know if AI can do a lot of ‘new’ things by the end of 2022, but I hope it can do better, she said, adding that it means it will be fairer, less prejudiced and more inclusive.
Samim Winniger, an independent AI researcher who previously worked for a major tech company, said he believes there are revelations about the use of machine learning models in financial markets, espionage and healthcare.
It will raise big questions about privacy, legality, ethics and economics, he told CNBC.
