By Yina Moe-Lange
Most agree that Artificial Intelligence has become a very hot topic and is changing our world a great deal. The applications of AI and its subsets are quickly gaining traction, and the question of when AI will take over human performance is being brought up constantly, but there seems to be no general consensus of the actual timeline.
A group from Oxford and Yale set out to try to answer the question, “When will AI exceed human performance?” They surveyed 352 researchers from the 2015 Neural Information Processing Systems (NIPS) and International Conference on Machine Learning (ICML) conferences, which are some of the top conferences for peer-reviewed research in Machine Learning. The main conclusion they came to was that AI would outperform humans in basic tasks quite soon, like 2024 for translating languages or driving a truck in 2027. And subsequently, the researchers believe there is a 50% chance that AI will outperform humans in all tasks in 45 years and in 120 years, all human jobs. But here is one thing to note! Beware of the large deviations in answers and the general lack of specific consensus on AI’s timeline.
The figure below shows the survey results of the question of when High-level Machine Intelligence (HLMI) will be achieved. Looking at the graph, you can see a wide range of answers, a sign that there maybe is no real consensus on this question.
This next figure shows the researchers’ answers to AI milestones. Again, the intervals here are spread out over many years. And more specifically, if we look at the last milestone on the list, the Go board game, the average prediction is about 12 years after 2016 as when an AI system will achieve the same level as a human. This we now know is wrong, as Deepmind’s AlphaGo recently proved to be the best in the world.
After this paper was published, many AI experts and researchers also chimed in with their opinions. Elon Musk noted on Twitter he thought it would be closer to 2030,
Probably closer to 2030 to 2040 imo. 2060 would be a linear extrapolation, but progress is exponential. https://t.co/e6gyOVcMZG
— Elon Musk (@elonmusk) June 6, 2017
While these experts may have answered the survey questions to the best of their ability, nobody can quite pinpoint the pace at which AI is progressing. AI predictions routinely underestimate reality because they are not taking into account that the field itself is growing exponentially. The brightest students are jumping into the field and the funding is rising dramatically. These factors naturally bring everything up ahead of schedule. One only needs to look at Google search trends to understand the pace that people are getting interested in various AI topics. The interest trend for Deep Learning has increased significantly, even just since the AI researchers were asked the questions about AI progress.
With more and more resources going towards AI, Machine Learning, Deep Learning and the likes, it will not be surprising if almost all of the predictions about the pace of AI are completely off. Also, here is an article where a bunch of smart people try to answer questions about the future of AI. Once again, the answer to the question, “When will we have computers as capable as the brain?” leads to answers ranging from 2029 to well over 100 years.
So, who is right?