Artificial Intelligence and other Emerging Technologies are gradually, but fundamentally changing the world we live in and this has become clearer than ever before during our ongoing COVID-19 crisis. During the week of 8 June, industry professionals, academics and leaders around the world gathered virtually for this year’s CogX event, the largest of its kind on AI and contemporary technology. All prepared to answer one major question: “How do we get the next 10 years right?”. It was an intriguing couple of days, and I’ve written down some thoughts around the main topics:
Up until this point, machine learning has mostly been about upskilling robots to do labour for us as humans. The ultimate goal is to make machines autonomous, and one of the big topics this year was how we can create an environment where humans and robots can co-exist while we move from teleoperation to autonomy. A step towards this is the concept of shared autonomy, which leverage the strengths of both the human and the machine, and in an ideal world, foster collaboration between human and robot.
There are multiple challenges holding back the development of AI today, and much of the research indicates that the main factor revolves around data, or lack thereof. In order to turn machines into autonomous workers, they must be trained, and this training often involves giving them thousands of examples. There are various reasons as to why this data is difficult to get. First, giving thousands of examples to a machine is a time-consuming endeavor, and the outsourcing of data labelling may not be an option due to increasing regulations around data privacy. Secondly, the current ways we are labelling data are not very efficient.
In order to effectively teach machines right and wrong, we need a paradigm shift that democratises machine learning, and we need the infrastructure and UI to facilitate this. Today, the subject matter experts are not the people responsible for developing and training AI models, but with the right tools they’ll be able to teach machines with fewer examples than we need today by focusing the attention to where it should be. As an example, a doctor may be able to teach a machine created for cancer screening quicker than the developer could, by directing the machine’s attention to where the cancer would most frequently occur.
Crowdsourcing is another hot topic, that with the right infrastructure, may help speed up the training of machines, and move towards autonomy more quickly. There are different takes on crowdsourcing in the context of AI, but common for all of them is that it includes the practice of using a large number of people to get work done. This use of crowdsourcing does not only have the potential to speed up the training process, but can also augment the intelligence. Four different interactions of collective intelligence were discussed at CogX which were:
- Machines working on data generated by people and sensors
- People and machines taking turns to solve problems together
- Solving tasks together at the same time
- Applying AI behind the scenes, connect knowledge and tasks in groups
A specific type of AI was particularly interesting, namely “Swarm AI”, which uses AI to connect people together modelled after natural swarms. This idea is rooted in what evolution has done naturally for millions of years, where nature forms systems in real time, with feedback loops and create optimal solutions as a collective. If birds, bees, and fish can get so much smarter by thinking together in swarms – why cannot we do this as humans? Such collective intelligence has the potential to amplify intelligence, swarms of medical doctors has reported to increase accuracy by 30%, whereas financial traders have been able to increase their forecasting accuracy by 25% working in groups of experts.
Most importantly, crowdsourcing, or collective intelligence has the potential to allow a wider audience participate in the AI debate to remove some of the inequalities that exist. We have already seen great examples of collectives working to uncover harms from algorithmic-decision making, Joy Buolamwini’s Gender Shades Project (part of the MIT Media Lab) is currently working to remove bias in image recognition to more accurately represent minority groups. Similarly, collective design can improve the AI by ensuring not only more diverse data sets, but more diverse voices at every stage of the AI lifecycle. In 2019, a research group at New York University released a study on the diversity crisis of AI and it highlighted the lack of diversity among the people who create artificial intelligence and in the data they use to train the machines. By allowing only a fraction of the population determine what is right and wrong, it poses serious risk of transferring problems of systematic inequality to machines.
It is therefore crucial that we as humans, before allowing machines to make decisions that will affect our lives, ensure that everyone in society is equally represented, and for that the crowds need to be diverse. Priya Lakhani, CEO of Century Tech is one of the visionaries trying to get AI out to the people, and one of tools she discussed is the free online course “Elements of AI”, which is specifically designed for people left out of technology discussions, with the aim of educating 1% of the entire Finnish population on AI. This is merely one example of what we can do to make further advances in the fairest ways possible. I for one, am intrigued by the promise of the technology of tomorrow and I look forward to seeing it evolve over the coming years.