Are We Already in an Artificial Intelligence Winter?
What developments are needed for AI to continue advancing?
Photo Source: Roger Montgomery
If you are a reader of Don’t Count Us Out Yet and also subscribe to TD Publishing’s Humanistic Artificial Intelligence Monthly Updates newsletter we started in November, you are aware of the way we teach artificial intelligence from a humanistic standpoint. We state how in order to understand AI and where progress is needed, one must start with the definition of human intelligence, which contains five parts: reality, memory, rational thinking, consciousness/emotions, and spirituality/religion and connectivity to nature. Once these are understood, one must figure out where AI tools are strong and where AI is weak relative to human intelligence. In some of these areas, we currently lack the tools needed to train AI that way.
Large generative languages or even multimodal approaches, which can use text, images and video, still don’t have the ability to create emotions or consciousness themselves. Currently, these tools just mimic any past data they scrape from. New, creative breakthrough thinking is simply not where it should be.
Let’s take an example. Suppose a new virus randomly appeared tomorrow with no current data on how to treat it. AI processes would be entirely useless until human intelligence gained knowledge about it, and only then AI would be able to scrape data about the virus. Therefore, when thinking about factors such as common sense (which some researchers are trying to replicate through Bayes’ theorem in mathematics, intended to describe the probability of an event based on prior knowledge, but aren’t close) or lateral thinking (a solution from something unrelated that might be tried here) without the right question from a human, artificial intelligence won’t even get close to a right answer.
In addition to these thoughts, there is a problem right now with overload of what to do, how to do it and where AI might go wrong. It just seems to us we aren’t even close to having an AI process do much more than stuff computer programs have already done, but in a quicker, faster and more efficient way. Take a look below at what we wrote in December regarding trying to stay up-to-date on what is important regarding artificial intelligence.
The first two articles below are excerpted from TD Publishing’s monthly newsletter regarding the overwhelmed information systems behind AI and where it is headed despite all the noise. If you would like to subscribe to TD Publishing’s monthly newsletter and stay up-to-date on advancements in artificial intelligence, please email jay@tdfactfind.com.
1. Are you kidding us? You expect us to do what to stay up-to-date with AI developments?
As expected when a year ends, many writers, blogs and so-called experts give you their best of reviews for staying up-to-date on certain areas. Artificial intelligence in particular, although a new area, had many reviews this year. We mentioned last month how overwhelming it was to consume the vast amount of information pouring in on new developments, what is important, and how to use AI and its tools. Sure enough, the surge of information at the end of the year was just another example of how much noise and time one would need to spend reading and listening to all the recommendations. Consider the following: Michael Spencer wrote about the top newsletters covering AI on Substack and came up with a list of 41 newsletters...
No way... I couldn’t read 41 even if I wanted to, let alone be able to digest what is good, what is noise or what is redundant on an ongoing basis. Add to that another newsletter’s 124 best tutorials to master AI and you have more than two months worth of full-time work just to stay current.
To us that is crazy, which is why TD Publishing’s monthly newsletter tries to give you the best of new ideas and thoughts in a 30-60 minute read. Yes, it will miss some of the big ideas, but it will keep you at the cutting edge without spending your whole life doing it. By the way, after reading this stuff for two months, you would be obsolete with all you missed.
That is what we wrote in December. Now, we have some more thoughts to add.
It seems to us that a number of obstacles are in place for AI to overcome before it can make a huge difference. Tools to help AI experience pain, pleasure or humor, or even know the difference between wisdom and intelligence, just aren’t there. However, in many cases these tools are essential for AI to be able to create a reality without us as good as we can now.
Yale Professor David Gelernter’s classic book written on consciousness in 2016, titled “The Tides of The Mind,” states “the role of emotion in thoughts our use of memory, the nature of understanding, the quality of consciousness…all change continuously through the day.” To us, that doesn’t sound anywhere close to how machines are constructed with semiconductors and on/off switches. There is plenty of work and new tools that need to be created for AI to truly grow up, and they just aren’t active yet. Deep learning was a huge advance that Jeffrey Hinton and his associates created in the last decade, but it doesn’t even address this issue that creates our intelligence.
Gary Marcus in his Substack newsletter, Marcus on AI, does a much better job than we could on covering some of the big obstacles in the way for OpenAI now.
Until some big new tools are developed, such as deep learning, we think the ability for AI to do anything more than just be a mirror of human data that exists is going to really slow down. AI can't expand our reality in a whole new way by just mimicking old data, or as Marcus states in the opening sentence of his book, “Rebooting AI,” which he wrote with Ernest Davis, “Since its earliest days, artificial intelligence has been long on promises, short on delivery.” We don’t need 41 so-called expertly written newsletters or 124 tutorials. Right now what we need are major breakthroughs. This might be the time to buy a coat for what is next in the AI world, as it is starting to get a bit frosty.
When I teach, I love to leave the class with three questions from what we talked about, and here are my three for all those “AI experts working in the industry.”
Will there ever be a killer app that makes it obvious AI is able to be humanlike? Or will AI just progress in specialist areas bit by bit?
When will AI be able to use mathematics, music, poetry, humor, critical thinking and writing to create something new (such as hip hop in music) instead of just mimicking what we have now?
Can AI really ever create thought like humans without switching to a biological mechanism that flows instead of an on/off electrical one?
We would love to hear your comments on this.
Best,
Craig for the Don’t Count Us Out Yet Team