What we are getting wrong by making AI humanlike

[edited using GPT4]

As we try to grasp the abstract concept of God in various religions, we often make it in our own image, making God human-like. It troubles me that we are trying to grasp the concept of a new kind of artificial intelligence by making it in our own image, human-like, and therefore limiting our understanding of its potential.

Imagining AI as human is comforting; it helps us fear it less and comprehend its intelligence. However, it also gives us the wrong mental model, distorts our expectations, and limits our imagination of its potential use cases. It is not uncommon to think of AI as an assistant, a computer that thinks faster, a robot that replaces human jobs, or a companion with emotional recognition that can actually talk. If we think of AI as human-like because it can talk, has eyes, and can recognize human emotions, then we might imagine that it can reason, have feelings, possess common sense, understand the world and its complexities, have desires, purpose, etc. However, we know that AI does not exhibit these qualities (not yet anyway), so it can be confusing for people who interact with AI to understand which human values it has and which it does not. Where is the line, or is there even one?

From current AI deployments, we see that the most effective ones are those that are reimagined, taking into account AI’s strengths and weaknesses. My favorite example is Amazon Robotics. Amazon Robotics did not replace humans by creating walking robots that pick up products in warehouses. Instead, they completely changed the game. They designed warehouses specifically for their robots, which don’t look like humans at all, and Amazon has redesigned the way it thinks about fulfillment by separating humans and robots. A similar use case is autonomous cars. The reason why we don’t have fully autonomous cars on the streets is that we are mixing humans and robots in one system, and humans can be extremely unpredictable agents. If we were to design things from the start and design streets purely for autonomous cars, we would all be riding in them right now.

Instead, we should think about designing AI that is human-centric, not human-like. Human-centric means that it is designed around human values, strengths, and limitations, while at the same time leveraging the strengths of the technology.

Related posts:

As an ex-captain and ex-manager of Celtic, Neil Lennon should 1,000,000% get grief at Ibrox. When Paul Hartley went back with Dundee he got grief and rightly so. But it is naive to suggest that what…

Quality is the notion that we want in everything. Whether it is about food, clothes or even diamonds, we seek good quality. Not just because it makes us look rich but because quality brings…

Consumer data paints a gloomy picture for the retail sector, not just in Britain, but across the developed world. Stores are quick to blame e-commerce for the death of bricks-and-mortar stores. But…