What is wrong with AI?
--
One of the most frequent questions that I’ve listened to talking to clients is, “Is that truly AI?”
When I hear this, I talk about what AI is, talk about how machine learning is being applied in the scenarios, and I conclude with:
Yes, it is AI!
But I’ve realized that my answer is wrong. What the clients understand by intelligence is something far deeper and more complex than what academia or industry has already achieved or will achieve in a near future. The main problem is not what is AI or what it can do, but what people expect when they listen to the word intelligence.
To get it in perspective, when humanity pursued flying, we got balloons, airships, airplanes, helicopters, machines that are equal or even better than the things they were inspired on; when we pursued swimming or diving, we got the canoe, boats, ships and submarines, gigantic monsters that can make blue whale look small.
In a nutshell, humanity is used to surpass expectancy when making artificial things based on natural ones.
At the top of an already really high standard, daily the media publishes articles about how amazing our auto-driven cars already are, and how bad our economy is gonna be when they replace all drivers jobs, how AI is killing all repetitive jobs. When any regular person, out of the AI area, sees an AI presentation expects nothing else than a high intelligence polyglot driver, capable of identifying cancer, recognizing anyone’s face, and winning millions of dollar in stock trades, doing all that between smashing world masters in GO and chess, simultaneously.
The truth is as far as Alpha Centauri is from the Earth. Although some systems can do some of the tasks mentioned, they are specialized systems and were trained for specific problems. To get a clear picture let's understand a little bit more about AI.
What is AI?
When most people think about AI, they imagine robots, androids that are gonna kill, or domain humanity.
But in the dictionary, artificial intelligence is defined as:
“A branch of computer science dealing with the simulation of intelligent behavior in computers.”
Examples of that can be visual and speech recognition, knowledge representation, planning, learning, natural language processing… So we can say that artificial intelligence is intelligence demonstrated by machines(a little bit more about intelligent behaviors in this other article)
Deep learning
The year 2015 was a mark point for AI. A system called AlphaGo defeated a world champion GO player, and It became headlines all over the world. After that, China built a 10-year plan to become the world’s leader in high tech, “The made in china 2025 ” plan. That event was the turning point for AI and the first time that Deep Learning got the world’s attention.
The new kid on the block, Deep Learning, had been applied to fields including computer vision, speech recognition, natural language processing, audio recognition, social network filtering, machine translation, bioinformatics, drug design, medical image analysis, material inspection, and of course, board game programs. In all areas, it has produced results comparable to and in some cases surpassing human expert performance.
The whole point about deep learning is that the AI specialist gets an enormous amount of data in a specific problem and trains a model on it, meaning that they ask the model to try to identify patterns on the data. After enough data, the system is able to get intrinsic patterns. The good point about it is that if you have enough data about a problem you can have an AI making predictions, analyzing, classifying, and operating in lots of problems.
As everything is not always rainbows and butterflies, deep learning has its own downfall too. Bias data can generate lots of problems like gender gaps in accuracy and racist analysis. Lots of work will be needed if you just need to change some aspects of the results. You always will need lots of data, and samples of all the scenarios.
At the begging of the hype, some researchers thought that it would be the path to a generic AI, but more and more people realized that it is a really good tool for specialist systems, but it is not the path to a generic intelligence.
How is it useful?
AI can have several fields of study and can be applied in a high variety of problems, from natural language processing to robotics.
New startups and big labs are applying AI for new products in almost all areas of knowledge. The simple association is that where there is data, AI can be on it. You can say that data is food for AI. Or like Robert Mercer, founder of Cambridge Analytica said “There is no data like mode data.”
And as hardware technology advances, it becomes more and more cost-effective to feed these babies.
It has already impacted humanity deeply. Maybe deeper than any other technology in the past, not because it has fundamentally changed everything, but because the changes are so intrinsic that it is difficult to be noticed, they are automating several of our decisions, making us easy to be manipulated and even dumber.
The first big impact was when it took the Internet and learned our personal preferences and basically built our online world in our image, serving us with content hand-picked for us. It is verifiable in every online aspect of our lives. Your search on google returns different things, depending on your history of search, location, and buying preferences. Your social media feeds are very different from other people, and it changes based on your reaction to the content, and the list goes on, Youtube suggestions, Amazon recommendations, Netflix personalization, etc.
The second big change, and less imperceptible was in Business. Companies have improved fraud detection, they are making smarter trades, improving supply chains, automatizing several inner processes from warehouses to services desks, to business intelligence. It makes the companies more competitive, innovative, and client-oriented.
The third big impact, that we are still living, some call it perception AI. Basically, it built another bridge of communication with machines, images and sound now are more than only bits in hard drives, it gained a whole new meaning.
Google now is capable of identifying faces on images and videos, and building memos that show our family reunions, our kids aging, and remember us of nice trips.
Alexa can play music, turn on and off electronics, buy and add stuff on lists, etc.
Trends
Following that idea is possible to map where AI has already changed our lives and based on the landscape of the last impacts we can think of what is coming next.
Our lives already were impacted by 3 major waves of AI. The first one when AI built our online world in our image. The second when companies improve their outcoming powering their solutions through IA and the third wave when AI invades our phones and homes, listening and talking to us digitizing our physical world around us through sensors and smart devices, building OMO(Online-merge-offline) environments.
The trend in business will keep growing in the next few years, and more and more companies all around the globe will be applying AI. Analyst's power will keep growing and hybrid working forces will enhance human capacities while keeping costs down, Chatbots and automation will keep pushing forward the number of clients and demands that each person can handle. Businesses are going to be able to customize more and more of their offers and marketing campaigns to a client level. Sensorial AI will also affect working places and customize user experiences in physical stores, like AmazonGO.
Online merged to offline will become more and more common and it will get more space in education, health, transportation, and even government.
A really interesting point of view in this AI world is put in the book IA-Super Powers by Kai-Fu Lee a Taiwanese-born American computer scientist, businessman, and writer. Success executive in silicon valley, who worked at Apple, Microsoft, and Google. Link to the book on Amazon here.
Conclusion
AI already changed our daily news, shopping, entertainment, business decisions, and even our way to interact with our homes. We know that it can repeat an action that it was trained for, but can’t make decisions or improvise if something changes. AI can perform repetitive tasks, but It can’t deal with deviations or irregularities in the goals they pursue.
All kinds of AI will get more and more integrated into day-to-day activities until it surpasses our expectations. But until there we will be enjoying the easy life that the specialist systems have already created for us.
Keep in touch by following me here on Medium.
