What artificial intelligence still can’t do
Artificial intelligence in the modern day is capable of miracles.
Yet today’s AI still has fundamental limitations. To make better use of resources and research efforts in the future, it’s helpful to take a step back and honestly examine the strengths and flaws of today’s AI. Promising work is already ongoing at the frontiers of the field in each of the areas listed below to make the next generation of artificial intelligence more high-performing and robust.
A man went to a restaurant. He ordered a steak. He left a big tip. When asked what the man ate in this scenario, a human would undoubtedly respond with a steak. Even today’s most advanced artificial intelligence has trouble with such cues. How is it possible? There is an almost unlimited number of facts about how the world works that humans learn through their daily lives. A person who is having a heavy meal at 7 p.m. will be less craving a second meal at 8 p.m. If I ask for milk, I’d rather have it delivered in a glass rather than a shoe.
The fact that humans form lasting mental representations of the objects, people, places, and other concepts that populate our world — what they’re like, how they behave, and what they can and cannot do — is the source of human “common sense.”
Continuous learning mechanism:
The usual AI development process is generally split into two stages: training and deployment.
An AI model learns to execute a task by feeding a static pre-existing dataset during training. The parameters of a model are fixed once the training phase is completed. The model is then deployed into production, where it creates new data insights.
Continuous learning in AI is limited by a phenomenon known as “catastrophic forgetting,” which explains why it has been so difficult to achieve thus far. In a word, catastrophic forgetting occurs when fresh information (new data) in a brain system interferes with or completely overwrites previous learnings (old data).
Cause and Effect:
To give a simple example, a machine learning model would have no trouble identifying that rooster's crow when the sun rises if given the correct data. However, it would be unable to determine whether the sun rises as a result of the rooster’s crowing or vice versa; indeed, it would be unable to understand the terms of this distinction.
We know that dropping a jar would shatter it, that drinking coffee will make us feel stimulated, and that exercising consistently will make us healthy. Causal reasoning is an important aspect of human intelligence that shapes how we make sense of and interact with our world. Developing AI that understands cause and effect remains a thorny, unsolved challenge.
Microsoft launched Tay, an AI personality, on Twitter in 2016. As a fun, interactive demonstration of Microsoft’s NLP technology, Tay was supposed to engage in online chats with Twitter users. It did not go as planned. Internet trolls had got Tay to tweet a wide range of offensive things in just a few hours, including “Hitler was right” and “I hate feminists and they should all die.” Tay like most AI systems today lacked any real conception of “right” and “wrong.” She did not grasp that what she was saying was unacceptable.
We may begin by defining particular rules that our AI systems should follow. In the case of Tay, this could be done by creating a list of insulting phrases and inappropriate topics and training the chatbot to avoid them at all costs.
How can we hope to create artificial intelligence systems that act ethically and have a moral compass that is similar to our own?
We don’t know, is the short response. The most promising field of research on this problem, however, focuses on developing AI that tries to figure out what people value based on how we behave and then aligns itself with those values.
As artificial intelligence becomes ubiquitous throughout society in the years ahead, this may well prove to be one of the most urgent technology problems we face. I am positive about the potential of a better society in which intelligent machines assist humanity to solve its greatest difficulties, improve the world, and make our daily lives easier.