Hello
This is your annual update on Artificial Intelligence (AI)!
Long read…but I promise it will be worth it
The Hype is Back
The AI hype died out during the COVID19 Pandemic, but now the “Hype” is back. An AI war is starting between companies like Microsoft, Google, and Facebook.
But Why?
Because the mega IT companies need to keep growing and make sure they can more money. So they have to start showing off what they are working on behind closed curtains. Maybe dust off some of the old off-the-shelf toys that have been setting in the basement for years.
But there is a major problem with corporate executives and AI.
Here is how I explain my stance on the corporate AI talk (and broadly any technology hype).
If a company created a great AI, something amazing. Then show it to me, show me how it works, let me use it for free, and open source it. Otherwise, it has ZERO value to me and to the world.
This sounds counterintuitive to traditional way American businesses operates. It place ultimate importance on Intellectual Property (IP) or No How.
However, AI is such an unique case, let me explain why.
If someone or a company creates an AI smarter than humans, wouldn’t they hide it from the public? Logically, they would just use it for themselves to gain advantage over competition.
If a company claims that they created an AI technology, but they don’t show how it works, and they don’t let people use it freely, and do not opensource it. Then what exactly is being shown to the public?
One possibility that this could be incomplete technology that is being hyped to make money to fund further AI development efforts, this has no real value people like you and me.
Another possibility is this hype is absolutely nothing. In this case, it could be simply a scam posing an AI as a viable technology, companies ask people to pay for a technology solely based on some vague AI advantage.
Scams are not new when it comes to AI technology. They actually date back to the 19th century with the Turk: A magical AI machine that plays chess. See below how it works.
So, how do we avoid getting Turked. Here is my tests for any company working on “AI”
Can you explain and show us how the AI works?
Can you opensource the AI?
Can you make us freely use this AI with no limits?
The Most Important Tech Human Will Ever Create
Survival is the most important thing for humanity. Some people may find a benefit to death due to old age as it renews and recycles societies, but this may be a coping mechanism to deal with the fact that death is currently inevitable.
Many people strive to enjoy life and do good in hopes have a great afterlife. Others over centuries have been trying to find an Elixir to eternal life. Until this day, billions are being privately spent on longevity research in hopes that humans can live forever and enjoy their wealth forever.
On another hand, some people fear that humanity need to solve the “death” problem because humans are not having enough babies 👶, AKA the population collapse doomsday crowd.
Wait, but how this whole spiel above is this related to AI?
Enter Google DeepMind Demis Hassabis strive to solve this ‘’small problem ‘’ 😀
While Google as a company sucks in commercializing their best AI to actually make human life better. There is a little corner of Google called DeepMind, which was a company bought by Google long time ago, DeepMind claims to work on cutting edge AI that they say is the best that ever existed. This AI system achieved super human skills in games like Chess and Go.
Once DeepMind proved that this AI system is great a couple of years ago they pivoted to real applications, starting with AlphaFold.
AlfaFold is a computer program that predicts protein folding with high accuracy with minimum effort, using AI. Why is this a big deal?
This protein folding problem would accelerate drug discovery, which ultimately helps medicine. Will the medical field and government use this to fight death? who knows. But what is important is what is next.
Demis Hassabis (46 years old) recently revealed that his ultimate goal from the pivot and AlphaFold is not for drug and pharma reasons, but for a more deep fundamental research reason.
The goal is to create an AI program that simulates a virtual human cell.
And with this virtual cell you can simulate all biological activities including aging, then you can simulate every possible way to treat any disease 🦠 or simulate every possible way to reverse aging 🧓👵.
If this is successful, it will be the holy grail. This will be a direct applicable benefit to humans that solves both old people’s problems, rich people’s problems, and population collapse problems.
It is also worth to note that Demis and DeepMind are smart folks, they know that the current tools and methods wouldn’t allow or enable an Artificial General Intelligence (AGI). That is why they think we should try to extend our lives as long as possible to come up with tools to build this AGI.
The Least Important but Exciting AI Tech
Stable Diffusion surfaced to my attention less than a year ago. It is an AI technology that produces pictures from text description. Let me explain why this is exciting. First, this has never been done before. Second, you can try to search the internet for a picture of something you want using text. Like a picture of a blue and pink spoon (see below). However, there are two problems with this:
You don’t own the rights to the picture you find on Google.
It is not super customizable.
Stable diffusion and similar tools (Dall-E and Midjourney) changed this game. You can produce any picture you want from text description, edit these pictures by talking to the computer, create images and art with no drawing skills. You want proof, see picture of blue and pink spoon coming from Google search versus generation by Stable Diffusion below.
The implications are very exciting because people use this in very creative ways, generating unlimited amount of images of anything people can imagine at very low coast, unlimited stock photos, unlimited art. You can generate HD pictures on your PC in 2 or 3 seconds using $1000 graphics card.
Why is this exciting? The next generation of this technology will allow you create videos (this is being tested now). This means that the entertainment industry can be easily personalized and customizable. Using AI, the plot for movies you will watch in the future could be created by you, you can have full control on what the characters do, this could be exiting to some, or boring to others. If people don’t like this create your own adventure media idea, movie creators can still use this to create unlimited amount of movies at very low cost.
For the Ad industry, Ads will be customized for you with your name in it and just the way you like ads. This can be the same for news, political speeches, or any consumable visual media.
Chat
Large Language Models (LLMs) are touted by many to be the next frontier for AI.
The good old days of AI had the Turing Test to determine if machine achieved intelligence. The test was named after Alan Turing who is considered the Father of Artificial Intelligence. Here is how it goes:
In the Turing test, a human evaluator would judge natural language conversations between a human and a machine designed to generate human-like responses. The human judge knows that one of the two partners in conversation is a machine, and all participants would be separated from one another. The conversation would be limited to a text-only channel, for example a piece of paper passing though a slit or a computer keyboard and screen. If the judge could not reliably tell the machine from the human, the machine would be said to have passed the test.
ChatGPT, a LLM computer program created by a company called OpenAI already passed the Turing for sometime now. There are thousands and thousands of pages on the internet that are already written by ChatGPT, and thousands of media and news articles published written by ChatGPT. You wouldn’t notice it.
In the business world, Microsoft saw this development and pledged to invest 10 billion USD in OpenAI to bring this functionality to its products and make a dime. It was a nice play by Microsoft’s CEO Satya Nadella who is a pro-CEO player making money buying LinkedIn, Skype, and Github. However, Satya’s bet backfired because the product didn’t work very will, but no big deal, I am pretty sure Microsoft will figure things out.
Zooming out, the real opportunity here is not a money makin. Last month, Steven Wolfram (a pretty smart scientist) described this AI opportunity in a very long article about ChatGPT; it appears that ChatGPT is actually a scientific discovery.
After some scrutiny and reflection, I tend to agree with Wolfram. This program should not work that easily, it seems that LLM unveiled that that the structure of language itself is computable.
This says something remarkable about the human brain and intelligence. It also demonstrates that knowledge contained in language itself can be compressed and modeled.
ChatGPT and Stable Diffusion are able to create images and text that appeals to the human brain. The sad and scary part about this discovery is something Wolfram barely mentioned in his analysis.
The reason the ChatGPT worked is that generating human appealing text is not as hard of a problem as we thought.
In my words I can paraphrase this as:
Humans are not as intelligent as they think they are.
This is not only humbling but also scary because it may offer an explanation for the Fermi paradox. There is no great filter, we are just not smart enough for other civilizations to interact with us. As Neil deGrasse Tyson puts it “maybe aliens looked at earth then said, we do not see any sign of intelligence here”
Here is something to think about: OpenAI engineers spent countless hours and the company spent 60 million dollars on training a 175 Billion parameter model/algorithm to create ChatGPT to generate human like text. Then Andrew Karpahy, a former Tesla AI engineer, created a similar program in 2 hours. Maybe humans are not as smart as they think.
Zooming in on tech companies, LLMs are causing Google to panic, the company as I mentioned before sucks in commercializing AI technology, it has a very hard time getting technology from the laboratory to consumers. The problem seemed to stem from the reality that big technology companies are not being run by founders anymore, but by MBA technologists, with a focus on product improvement and money making.
Moreover, political division and lack of unity led to the rise of employee activism in companies like Google and Apple with their efforts shifting to social issues rather than AI technology breakthroughs. In a nutshell, Google is Too Sensitive and Too MBA to create something like ChatGPT.
Apple, are you okay? Apple still makes good products, the spirit of Steve Jobs is probably touring Apple’s HQ and kicking employees in the ass to keep the Art/Innovation theme alive in this company. However, Siri is pathetic, and Apple has no cutting edge AI technology that compares to Google, Facebook, Microsoft or OpenAI. Unfortunately, Apple is only about art and great products, not about AI. Apple is the prime candidate to bring AI technology to the benefit of humanity, instead they are bringing us another season of Ted Lasso.
I can definitely dig into all the exciting details of how ChatGPT works for my readers but I believe the public has no interest in learning this, it is just nerds like that stuff. But if you are interested to know, drop me a comment or email and we can chat.
The Singularity
In the 1950’s, Neuman and Ulam (two scientists) coined the term “singularity” to describe a context related to the acceleration of human technological advancement. The meaning of singularity has changed since then, it now refers to a point in time where technological advancement will become unhinged and reach an escape velocity where it is unstoppable.
Later the singularity concept was used in the context of AI, where if humans created an AI system that is more intelligent than humans any general task, the singularity would be reached.
So did we reach ultimate machine intelligence?
Now that we know that ChatGPT and others can pass the Turing test, but does it mean that we are close to this singularity?
My current opinion is that it is likely that humans will not even observe that we reached singularity. If it happens, it will be a headline on the news. This news is going to die to two weeks. I believe that the two main reasons for this are:
We are not as smart as we think, see above
We have been expecting this for a very long time so when it arrives we will say “we told you so”
What is funny about this logic above is: the singularity may already have happened. But lets not go there today.
So humans are pushing hard forward towards the singularity. But unfortunately all these AI systems are not successful in achieving the singularity as we defined it, today’s AI systems do not work on any problem but only specific ones. The meme below (although not very accurate) explains a version of AI (heuristics), this AI is simply explained as follows:
if this happens, Mr. AI please do this
if this happens then do this, and so on.
You see it now? The generalization problem is the core problem to be solved. If solved the singularity can happen.
The question is, do we even know what would it take to build a system that can achieve AGI. I believe we do, but the majority of AI researchers are not using the guidebook. I wrote about this last year but here go again.
Here are the features of architecture that needs to be built to achieve human intelligence level and beyond:
Real time connected system that senses the real world (vision, sound, touch, smell and more)
Continuous learning of real world, new information changes the system
System learns and updates based on distributed discrete models
A methodology to reach consensus between models to reach and make decisions (for example voting)
In summary, here is your annual update on AI:
The AI hype is back
The best case is that we will use it to live forever
The exciting case is that it will entertain us beyond imagination
The moonshot case is achieving the singularity, only if we find the right pathway
So what do you think? Leave me a comment of email me with your thoughts.
If you want to dig deeper, here are some further readings and media to check out: