The Algorithm changed again? #ChooseToChallenge the daily interaction with AI technology
Summary written by Buvana Ganesh, Catherine Lalanne, Iva Simon-Bubalo
Women In AI Ireland, Microsoft Ireland DreamSpace and the Confirm SFI Research Centre for Smart Manufacturing presented this live virtual WaiTALK titled “Choose to challenge: The Future of AI and You” designed for non-experts from secondary school students all the way to professionals, on International Women’s Day 2021, to familiarize them with the basics of AI.
This WaiTALK introduced the audience to the various applications of AI technology in the services they use daily, clarifying the associated potential, limitations, perceptions, and risks of new technologies in an engaging way.
What is AI and how has it become part of our everyday life?
Amanda Jolliffe leads the teaching & learning execution and strategy of DreamSpace with the Microsoft Ireland Education team. DreamSpace is the STEM education venue in Ireland and works very closely with school communities to raise awareness of digital technologies and to increase the opportunities for people from less advantaged backgrounds, to ensure that we do have more diversity in STEM and in things like AI.
In her interactive talk with the audience, Amanda Jolliffe discussed technology behind tasks like: Identify a person through their facial contours; process the language you speak and provide you with an answer; given your behavior as a user, recommend video you may like to watch — prompting answers around Facial Recognition, VoiceAssistant with Natural Language Processing and YouTube / Netflix / TikTok recommender systems.
Daily examples of user interaction with AI technology online
To help the audience recognise and learn about the situations in which they are interacting with AI technologies, Amanda Jolliffe performed two live demos:
Facial Recognition Demo: Amanda uploaded an official photo of Mary McAleese to a demo which recognized Mary’s hair color, gender, age (incorrect by a few years!) and Facial features coordinates.
Microsoft Azure Health Bot Demo where she went through a fictional conversation with a bot guiding her to best ways to manage her headache.
However, any AI, like the demos above, is based on available data : incomplete, biased data can result in faulty decisions. It is thus essential to increase diversity in data gathering and AI professionals. More about Microsoft Responsible AI driven by ethical principles — https://www.microsoft.com/en-us/ai/responsible-ai
Time to weigh in on the debate rounded this very enjoyable and engaging session!
- AI should fact-check all conversations had over the Internet.
- Online shops should ship products to us before we even know we need them.
- Social Media Sites should only connect you with like-minded people.
Public perception of AI and myth-busting
Sean O’Brien is the Education, Public Engagement & Training manager with CONFIRM, the Science Foundation Ireland funded research centre for smart manufacturing.
In his talk, Sean busts three most mainstream AI myths: “AI will take our jobs”, “AI will overtake and outpace human intelligence”, and “AI will wipe out humans”.
“AI will take our jobs” myth. In the past, during times of the beginning of industrial revolutions, the general public had adopted the same way of thinking and fear of technological advancement like we observe today around AI technology. People created misconceptions about telephones transmitting “evil spirits” directly into their heads, and trains that would make passengers suffocate because of the speed. With the mass adoption of the new technology over the course of decades, usually, these fears become trivial as people interact with the technology and form experiences.
“AI will overtake and outpace human intelligence” myth. Artificial Intelligence, although inspired by the workings of the human brain and reasoning, is still very far from possessing human level intelligence and consciousness. There is nothing “conscious” about computational processes that enable machine decision-making. Think of it as a simulated reasoning with no subjective human-like internal experience from the machine’s side. Machines are currently good at tasks that they are designed to do, and perform very poorly in open-ended situations humans can easily solve.
“AI will wipe out humans” myth. Sean introduces the third myth with a couple of compelling, even philosophical questions: Can AI be inherently evil? Or is it actually programmed to become so? If the AI is following a mission for a greater good, is it actually evil or do we perceive it so? Although these are open questions, with the current state of AI technology these scenarios are highly unlikely. They were primarily introduced to the public through movies and novels. There is no scientific evidence suggesting that we are headed in this direction.
As a result, scientific communities and expert opinions in the past and even now seem to have less influence in shaping the public opinion than the movies and dramatic scenarios.
How AI knows… what you want to watch next
Begüm Genç, member of WAI and a Postdoctoral fellow in the Confirm Centre, then spoke about Recommender systems and their prevalence in our everyday life. She started off by talking about how AI has been around since the 1950s, but has progressed faster now because of the abundance of data, multidisciplinary collaborations, etc. Thus, AI has now evolved into different types like Natural Language Processing, Machine Learning, optimisation, etc.
Begüm then delved into the topic by asking the audience which of Netflix, Spotify, Tiktok, Youtube, Amazon, Instagram people they use. The commonality between all the apps is that they all recommend things that they specialise in, to us. But these can be biased based on the data or the algorithmic, as in the case of Tiktok where a user identifies that the recommendations he gets are based on the similarity in profile pictures to accounts he had already liked.
“(Ethics researchers) question about how to make AI do the right thing. The developers who have AI ethics in mind often as themselves ‘should I code this’ rather than ‘can I code this.”
When there is not enough information about someone who is just starting to use the app, this lack of data for recommendations is called the Cold start problem. On the other hand, Apps like Spotify can become complicated because of the excess information from sub-profile creation, categorisation, etc. Therefore, user engagement is monitored to cater to specific needs. Reinforcement learning is a particular way an AI learns by rewarding good decisions of the algorithm and penalising when bad decisions. This can be seen in the OpenAI demo of Hide and seek game with 2 hiders and 2 seekers and how this rewards mechanism makes the seekers progress from random movement to coordination to using available props, etc.
More diversity and women in AI
Therefore research in AI is very important. She asks where women stand in this AI race. We find that only 22% of the workforce, 12% of leading ML researchers are women. To change this narrative and show support for women so that they don’t drop out or move past this arena. Begüm showcased prominent women working in this field throughout the presentation, like Timnit Gebru ex Google AI ethics research, Mounia Lalmas in Spotify User engagement, Daniela Rus in MIT CSAIL and the importance of their work in AI.
Women in AI is the first global community of women in Artificial Intelligence with over 3,000 members in 90 countries. We aim at shaping gender-inclusive AI that benefits global society by increasing female representation and participation in AI. Our global platform empowers female AI-visionaries and practitioners by providing access to educational resources, networking opportunities, and hands-on support, while accelerating a community of global pioneers in the field of Artificial Intelligence.
Follow us on:
