Albeit being at the forefront of technology, there are dozens of challenges of AI that it has to conquer to make it completely independent of human assistance.
Have you heard of J.A.R.V.I.S.? This is one of the most popular Artificial Intelligence (AI) assistants to Tony Stark, aka Ironman. Although J.A.R.V.I.S. is a fictional AI, we already interact with many similar applications of AI, thanks to the rapid evolution of the technology. There’s a set of people who would love to explore more about AI, while a different set is against the idea as they believe it might “take over” the world in the near future.
The morality of AI taking over humans is a separate matter. For now, it is important to know that even though AI can perform multiple tasks beyond one’s expectations, it cannot do everything. There are several challenges of AI that have not been addressed yet.
AI Cannot Make Moral Decisions
AI can abide by a given set of laws and rules far better than a human. However, the technology is lacking when there’s a moral dilemma involved. A driverless AI-powered car is a perfect example of this, wherein the AI technology has to decide whether to steer in a direction that would lead to potentially colliding with a barrier to save the pedestrians or to hit them and save the passengers of the car. The chances of casualties in both cases are high and AI is incapable of taking all the factors into consideration, unlike a human driver in a similar situation.
AI Doesn’t Have Common Sense (Yet!)
Humans operate on a bank of knowledge called “common sense,” which sets us apart from AI. AI has to be fed with rules on how to hold a bottle or perform a particular task, whereas a human can do the same without making any errors. AI is fed with data and using the database, it can return correct or wildly incorrect results.
For instance, Kevin Lacker, a Silicon Valley entrepreneur, asked an OpenAI language model GPT-3, “Which is heavier, a toaster or a pencil”? To a human, it is common sense that a toaster is a right answer. However, GPT-3 pointed out that the pencil is heavier as it didn’t have much data to relate and produce the required results.
AI Cannot Learn and Adapt
Humans have a tendency to learn and adapt in real-time. This is called continual learning. On the contrary, AI goes through training and deployment phases where researchers would train it on a specific dataset and deploy it. However, unlike humans, AI can’t adapt to a new dataset without training over a ginormous dataset, which is a time-consuming process.
A researcher would anticipate scenarios and prescribe rules and laws that AI could follow to respond to a particular situation. However, this is one of the biggest challenges of AI, at least at the moment. However, work is being done to instill AI with the capability to dynamically accomplish continual learning.
AI is far from taking over the world because we have just scratched the surface and there’s a lot to discover and deploy yet. AI lacks creativity, empathy, sympathy, cannot make moral decisions, doesn’t understand cause and effects, and these are just a few challenges of AI in the modern world. However, researchers are continually working to overcome these barriers and create a technology that can replicate human effort and be error-free.
