There’s no doubt that artificial intelligence is evolving at a breakneck pace, and 2019 will likely bring many more new, unexpected leaps forward in further developing machines with true, human-like general artificial intelligence. We know we’re not quite there yet, but with an abundance of AI-related stories out there, it can be hard to follow the threads of where we’ve gone in the past year, in order to know where we are going in the future. With that said, here are some highlights from the past year in the ever-growing world of artificial intelligence and machine learning.
1. Mind-Reading AI
One of the more interesting developments this past year is the notion that AI might be used someday to “read” your thoughts. While that can be an unsettling idea at first glance, the flip side is that such technologies might help those with disabilities communicate or see better, or even help improve image search (imagine finding a particular image merely by visualizing it). To demonstrate the possibilities, one team from Kyoto University, Japan developed a reconstruction algorithm that is able to “decode” and optimize complex visual information from the brain waves of someone looking at a particular object — taking us one step closer to machines that can accurately read your mind.
2. Machines with “Visual Foresight”
Deep learning is a particular subfield of machine learning that is inspired by the way biological brains are structured and function, and aims to develop better artificial neural networks that would underlie adaptable machines that can learn and think more like humans. Earlier in the year, researchers over at University of California Berkeley’s Artificial Intelligence Research lab (BAIR) created a machine that’s capable of visualizing its immediate future, using what they call “visual foresight.” Inspired by the way human babies tend to experiment and manipulate their environment in order to learn and then apply those lessons to new, unknown situations, it’s something that humans take for granted, but is actually difficult for machines to master. Such research could pave the way for a kind of “visual imagination” for machines, allowing them to autonomously interact with their surroundings.
3. Cooperative Machines with Social Skills
One might think that cooperation and other social skills are the sole domain of humans, but as recent experiments show, machines can also be endowed with an artificial set of social skills that would enable them to collaborate with both other kinds of machines, as well as humans. In creating such an algorithm for this artificial set of cooperative social skills, one of the researchers on an international team noted: “The end goal is that we understand the mathematics behind cooperation with people and what attributes artificial intelligence needs to develop social skills. AI needs to be able to respond to us and articulate what it’s doing. It has to be able to interact with other people.”
4. AI That Learns from Its Mistakes
Nothing makes us potentially more human than the ability to learn from our mistakes. One might find something similar in machines in what’s called reinforcement learning, but as researchers over at OpenAI point out, designing the reward systems behind reinforcement learning models can become quite complicated, and may actually discourage machines from exploring possibilities beyond the tasked goal. Instead, this team proposes an open-source alternative they call Hindsight Experience Replay (HER).
5. Self-Replicating AI
The ability to pass on successful traits is a defining characteristic of biological organisms. Earlier this year, two researchers from Columbia University found a way to apply this principle to artificially intelligent systems — creating self-replicating neural networks called “quines.” The idea of self-replicating, self-evolving AI that can automatically take on the most successful traits of previous generations is a pretty tantalizing one, with lots of potentially useful applications.
6. Cultural Bias in AI
The infallibility of the dispassionate machine is an easy fallacy to buy into. But as experts in the field are showing, there are plenty of hidden cultural and gender biases in our algorithms, one that can affect people for life when these algorithms are used in automated decision-making systems, such as those utilized in human resource departments or the criminal justice system. As we forge ahead with AI in 2019, it will be critical that we find ways to tackle this phenomenon of algorithmic bias, so that long-standing prejudices and social injustices aren’t further perpetuated through our machines.
Hear more: Cultural Bias in Artificial Intelligence
7. AI-Assisted Automation
Of course, no discussion on AI can detour around unavoidable fact that AI is helping to further automate all kinds of industries and professions, whether it might be in manufacturing, distribution or even white-collar domains like finance. While the prospect of human workers losing their jobs to ever-smarter AI systems seems like a bleak one, there are some positives to look forward to, such as using AI to help automate the prediction of new and dangerous drug interactions or automating game design — saving lives and potentially making it more entertaining.
Image: Franck V. on Unsplash