On this page
Building an artificial brain is hard because scientists still do not fully understand how human cognition works. Researchers are therefore borrowing ideas from human cognitive abilities, such as attention, thinking and perception, to guide the design of AI systems.
When AI first appeared, many people assumed that building a model that mimics humans would be simple. It has turned out to be far harder, and the goal of machines with human-level intellectual abilities remains unmet.
Key takeaways
- The human brain is only partly understood, which makes reproducing its cognitive abilities difficult.
- Neural networks are the main tool for modeling interaction between humans and machines.
- Attention, thinking and perception are three human abilities that inspire AI research.
- Machines learn to perceive by being shown large amounts of similar data.
- Human-machine collaboration already exists in assistants such as Watson, Siri and Cortana.
Why is it so hard to build a machine that works like the human brain?
Part of the problem is that researchers are still working out how human mental abilities function. Everyday routines may look simple, yet the brain performs complex cognitive work behind them. Scientists have explained some of its workings, while others remain a mystery.
That mystery is also why the brain attracts researchers as a model for machine intelligence. Reproducing cognitive abilities in a deep neural network is not easy, and it remains one of the hardest goals in AI.
The human brain also has a long evolutionary history. It developed in response to survival pressures, curiosity and the demands of nature. People extend their abilities with mathematics, algorithms, computational methods and statistics, which speeds up scientific discovery. Machines do not learn in the same way. Often, instead of letting a machine learn, engineers encode algorithms that control it.
What is the state of human-machine collaboration?
Collaboration between people and machines already exists in products such as Watson, Siri and Cortana. Neural networks are the main way researchers copy interactions between humans and machines. As they improve, other human cognitive abilities become more relevant to AI.
The original article argued that for AI to have a transformative effect, artificial neural networks need to be built on a foundation of natural human intelligence. More AI researchers are implementing algorithms inspired by specific cognitive mechanisms of the brain, and the original reported very promising results.
Interest has also grown in brain-computer interfaces. The original article named Elon Musk’s Neuralink and Bryan Johnson’s Kernel as startups that have intensified the debate by pursuing enhanced machine abilities through interlinking. Public curiosity about what a machine with every human cognitive ability would look like sits alongside fears of a bleak future.
Which human cognitive abilities inspire AI?
| Ability | Description |
|---|---|
| Attention | The ability to focus on one thing while ignoring distractions, and to shift focus when needed. |
| Thinking | Drawing on memory, information processing and recognition to reach solutions quickly. |
| Perception | Recognizing objects and sounds across different conditions. |
Attention
Humans can concentrate on one thing for a long time while tuning out their surroundings. They can also redirect attention when something new demands it. Researchers draw on this ability in deep learning, including work on convolutional neural networks (CNN) and deep generative models.
Thinking
People rarely spend much effort on finding solutions, because memory, information processing, and speech and object recognition work together in the background. AI researchers try to imitate this by building theoretical models of the brain. Those models draw on studies across disciplines of vision, movement, sensory control and learning.
Perception
Recognizing an object or a voice is difficult for machines. A model must be trained on many similar examples. For instance, after being shown many pictures of a monkey, a model can learn to recognize one. The mechanism behind the machine mind is therefore a reproduction of how the brain learns. It helps machines identify an object in different situations and under different lighting conditions.
Frequently asked questions
Why is it difficult to replicate the human brain in AI?
Scientists have not fully mapped how the brain produces cognition. In addition, the brain evolved over a very long time, and machines do not acquire abilities in the same way.
Which neural networks are mentioned as drawing on attention?
The original article links the brain’s attention mechanism to convolutional neural networks and deep generative models built with deep learning techniques.
How do machines learn to recognize objects?
They are trained on many examples of the same thing. After seeing enough images, a model can recognize the object, even in different conditions.
Do AI assistants already collaborate with humans?
Yes. Watson, Siri and Cortana are cited as examples of human-machine collaboration in use today.
Conclusion
Machines that match human cognition remain a distant goal, partly because the brain is still not fully understood. Studying attention, thinking and perception gives AI researchers practical models to build on. Progress will probably depend on how well neural networks capture these abilities.
