AI is certainly artificial, but not intelligent. Will it be?

What does " #artificialintelligence" do?

AI is certainly artificial, but not intelligent. Will it be?

Post No. 9

The reason you have not seen my blog in a couple of weeks is that I was traveling in Europe - England and France. England and Europe remain great gems. My wife crossed London with over 20,000 steps per day and together we toured the Normandy area of France. Visiting Omaha Beach and the American Cemetery sure leaves an impression.

We then spent a couple of days in Paris, and walked the gardens of Versailles, and enjoyed the great food.

Paris is now one big traffic jam, pretty much everywhere you go and most times of the day. This made me wonder, do we have to wait until most cars on the road will be electric and smart?

Can’t we solve traffic, which is a problem in many metropolitan areas, sooner than once all automobiles have been converted?

Of course, we have Google Maps, and the associated Waze. The Uber drivers seem to like Waze better, yet they frantically keep updating Waze during the drive to identify better routes, and yet we still seem to be stuck in traffic at every turn. I believe this is because when Waze provides alternative routes, it does so for many other drivers and based on the then current situation. This creates mini-clogs, as many drivers now divert to the alternative route, and it then becomes jammed.

So, we might be able to apply Deep Learning and Generative Adversarial Networks (GAN) to develop capabilities for also forecasting how traffic will behave, and then leveraging the broader network of roads, and not just highways, for alleviating traffic jams.

Afterall, traffic is a physics and math problem - flow and numbers.

And this brings me to the key point - AI today is a very powerful tool to handle large sets of numbers and data. It has immense capabilities to help us understand the world on a macro level vs. our own understanding, which is usually limited to more local situations and scenarios. We evolved to interact with a small set of connections - a few dozen - not with thousands, millions, and billions. However, computers, and computations are quite content with huge sets of data and highly repetitive tasks - and I am aware that I am anthropomorphizing here.

Eventually, it is the collaboration, and work that humans can accomplish together with machines / computers that will create value for everyone. From all I am reading and learning, I do not believe that the current artificial intelligence is close to the capability of having sentience and what we would consider thinking.

Even the highly sophisticated GPT-3, DALL-E 2, or Google’s LaMDA are extremely well adept at very specific tasks - verbal or image processing - yet this is based on internalizing an unimaginable number of words, sentences, and pictures, and evaluating patterns, then applying them to new problems posed to the algorithm. This may give the illusion that there is some “thinking” happening in the background, and provide some sense of sentience. However, this is only an illusion.

Recently, we hear of the Google engineer who claimed LaMDA was sentient. We also saw articles related to Google’s AI passing the famous Touring Imitation Game test. However, I believe that this simply demonstrated that the test, proposed in the 60's, is no longer valid.

So, we should focus on leveraging the great power and value that our current AI has to offer, while paying attention to the usual MLOps best practices, and paying close attention to maintaining our ability to explain what the AI algorithms are doing - “...with great power comes great responsibility.”?

In the next posts, I intend to delve deeper into the concept of “superminds” and the aspect of model drift that eventually impacts all AI systems.