Here, anyone can read, write, and share science.

Try it for free. No registration required.

The Link Between Neuroscience and Artificial Intelligence

Although considered a relatively new field of study, neuroscience is already playing a huge role in current technological innovation, especially in artificial intelligence (AI). Before attempting to create a robot with the ability to mimic human intelligence, a thorough understanding of the human brain is crucial. The first research into and construction of artificial neural networks began in the 1940’s, leading to ideas and proposals in which neurons could “learn” using feedback, or encode statistics. Soon after, the backpropagation algorithm gave multi-layer networks the ability to learn [2]. The two fields have continued to intertwine since, allowing for increased interdisciplinary collaboration.

Discoveries of both complex and simple brain mechanisms have accelerated the development of artificial intelligence. For instance, British AI company DeepMind made use of the neuronal phenomenon of “replay” in their deep-Q network (DQN) algorithm. Upon observing the brains of resting rats who had previously run through a maze, scientists discovered the same sequence of neuronal activity that had been produced during their active period took place in the resting period. In doing so, their brains worked to optimize behavior in the future by learning from mistakes made during the day. If this experience replay was disturbed, the rats performed poorly on tasks that were previously simple for them. DQN was able to improve performance in Atari 2600 games using experience replay to analyze data offline and learn using success and failures of the past [1]. Another example worth noting uses ‘working memory’, or the ability of the human mind to actively store and access information. The basic framework involves an executive central controller and separate memory buffers. In long-short-term-memory (LSTM) networks, this framework is paralleled: information is left in a fixed active state, where it is maintained until it requires the appropriate output [2].

Recent findings have proved that AI is progressing rapidly and is functioning in a way comparable to a human brain. In January of 2019, Google’s AlphaStar AI beat 10 out of 11 pro players at StarCraft II, a strategy game requiring quick decision-making reflexes [4]. Though many fear the imminent threat AI poses to the job market, especially white-collar jobs, there are numerous advantages of incorporating artificial intelligence into modern society. Recent research published by Dr. Jae Ho Sohn, a resident in the Department of Radiology and Biomedical Imaging at University of California, San Francisco, describes early-stage Alzheimer’s disease detection via application of a machine-learning algorithm to PET scans. After being trained on over 1900 scans, the algorithm correctly identified 92% (first dataset) and 98% (second dataset) of patients with Alzheimer’s, predicting the disease six years prior to the patients’ final diagnoses [3]. Alzheimer’s, often diagnosed too late for viable treatment, can now potentially be detected six years in advance.

`Although still in its early stages, artificial intelligence is a promising technological development, and by using the human brain as a model, programmers are able to take algorithms to new heights. Ultimately, collaboration between multiple fields of study has almost always brought triumphant results to the scientific and global community, and the cross between neuroscience and AI is no exception.

References

[1] Summerfield, Christopher, et al. “AI and Neuroscience: A Virtuous Circle.” DeepMind, 2 Aug. 2017, deepmind.com/blog/ai-and-neuroscience-virtuous-circle/.

[2] Hassabis, Demis, et al. “Neuroscience-Inspired Artificial Intelligence.” Neuron, vol. 95, no. 2, 19 July 2017, pp. 245–258., doi:https://doi.org/10.1016/j.neuron.2017.06.011.

[3] “Artificial Intelligence Can Detect Alzheimer’s in Brain Scans Six Years Before Diagnosis.” Neuroscience News, 4 Jan. 2019, neurosciencenews.com/ai-alzheimers-detection-10427/.

[4] Piper, Kelsey. “StarCraft Is a Deep, Complicated War Strategy Game. Google’s AlphaStar AI Crushed It.” Vox.com, Vox Media, 25 Jan. 2019, www.vox.com/future-perfect/2019/1/24/18196177/ai-artificial-intelligence-google-deepmind-starcraft-game.

 Edited by:
Mehek Dedhia