The Long Quest For Artificial Intelligence

Ever since computers were invented to automate some manual tasks, there has been a quest to mimic the way the human brain works — to create some kind of human intelligence using computers.

The effort had a long gestation period of about five or six decades. As a result, artificial intelligence, or AI, as it is known, dominates various aspects of our day-to-day lives.

AI has developed in different directions. While proving to be a boon overall, it is stoking fears of surpassing human intelligence.

For a long time, AI was always around the corner, like nuclear fusion has been for several decades, and even is today. Along the way, we reached a stage where AI became a distinct possibility, and the best we could hope for was fuzzy logic, which did make its way into gadgets like washing machines to optimise their performance.

Two things happened that made current AI a reality.

First is the progression of Moore’s law, which resulted in computer chips doubling in performance every 18 months. As a result, we now have the most powerful computer in our pockets in the form of our smartphones, which are orders of magnitude more powerful than the ones which guided the Apollo spacecraft to the Moon.

The second development is collection of huge amounts of data via the technology platforms: Facebook, Google, Twitter, Microsoft, Amazon, Apple, IBM, Tencent, Alibaba, and others.

Data has been collected to indicate the preferences and activities of billions of people in the world — all this in the name of offering freebies in terms of allowing free search and communication abilities. This is akin to politicians garnering votes by giving away cheap gifts. A lot of data is also being collected by the digital payment platforms.

Armed with these two developments, there is enough computing power and data available to the technology companies to look for hidden patterns within the data, which would not have been humanly possible to do previously.

Thus, we have "big data" and the software algorithms known as "deep neural networks."

Regardless of the buzz words, it led to recommendations of books and other products on e-commerce platforms, in filtering out search data and topics for users to consume, identification and discarding of spam emails, and so on.

Translation between languages has been another great success story for AI.

While on a recent visit to Panama, we used the phone to type in an English phrase and it translated to Spanish, which we then showed to the local folks. They then typed in Spanish responses on their phones and translated to English and showed us their phones. This strategy came to my rescue whenever my broken Spanish did not work.

Assistants like Amazon's Alexa and Apple's Siri started to recognise human language and respond back. Video recognition is leading to self-driving cars. More the humans interact with AI, the more it learns, like human babies, and the more it grows. It needs computing power and lots of data as key ingredients. But it is still a human-initiated exercise. The AI is not acting by itself.

Chess and Go are strategy games based on a lot of permutations and combinations, while Jeopardy is a knowledge-based game. AI has now reached a stage in this process of learning the rules of the game by itself, instead of being fed by humans. In a technique called "reinforcement learning," AI learns by trial and error. AI has learnt new games by itself.

Other than capturing data from customers, there are a lot of sensors in various types of machinery and CCTV cameras all over. The data gathered from these sources, coupled with cheap storage devices, means that we are drowning in data.

The data management software is freeware in most of the cases. The data can be stored and processed in the cloud without owning any piece of hardware or software. This is a bonanza for AI, which can predict when a machine will break down or is in need of parts replacement. Or try to recognise a suspicious personality from hundreds of CCTVs footage.

AI is being used in diverse applications like drug discovery in the pharma industry — to find the right chemical molecule from millions of combinations to fight a particular disease. It is also being used to analyze and decode the Sumerian script — the ancient writing thousands of years old from Mesopotamia.

The most recent applications to capture the headlines are ChatGPT and DALL-E. Both of them are from OpenAI.

ChatGPT is a large language model which can analyse millions of documents and summarise on a particular topic or an issue or a question.

Similarly, DALL-E can draw a picture based on clues supplied via text. It's like an artist working for the police who can draw the picture of a suspect based on the description supplied by witnesses.

As a result, Microsoft has invested $10 billion into OpenAI to get a headstart on this technology and use it in their products.

Both these tools are freely available for anyone to play around with. Essays are being written and pictures generated by lay people with no experience in doing so. I make the claim that I wrote this article without the help of any of the tools out there.

Such tools are part of Generative AI, leading towards what is called AGI (artificial general intelligence).

All the current AI capabilities of interpreting data and coming to conclusions is more like a domain-specific AI. The domains can be drug discovery or language transalation. It cannot be expected to solve general problems in another domain or across domains.

AGI is far out into the future. Generative AI is a possible bridge towards that goal.

Regarding the feared dangers posed by AI, I don’t see a possibility of AI overtaking humans or human intelligence, even if AGI becomes a reality and comes to dominate our lives.

AI will grow while humans will evolve to work with AI in a collaborative way. Technological advances will gallop with convergence of multiple technologies, and human lives will be different in the future, but not as slaves or the haunted ones.