In 2019, Unit8 announced our partnership with Digital Switzerland, a Swiss-wide and multi-stakeholders initiative with more than 150 members with the objective to contribute to the mission of “Making Switzerland a Leading Digital Innovation Hub”. As an active contributor of this network, it was obvious to us that we would participate in the “Digital Days”, the organisation’s flagship annual event, from November 1–3, 2020.
Unit8’s mission is to help Swiss and international companies accelerate the adoption of AI by working with our customers to solve real-life problems with software and advanced analytics. It is also our responsibility to demystify the “black box” of AI and make our work transparent and easy to understand. We want to demonstrate what AI can do today and help our partners and the general public realize the potential it holds for the future.
Now in its fourth year, Digital Days aims at making digitisation a tangible experience and fostering dialogue around the digital future of Switzerland. Therefore, we approached this year’s edition with the theme “Demystifying AI” and asked our team to contribute by sharing their expertise in a pragmatic and easy to understand way.
Our team led the organisation of two “Learning Labs,” a session format created by Digital Switzerland with the objective of teaching a specific digital skill in a 45-minute online webinar.
In an era in which “fake news” generates doubts and fear about whom/what to trust, this Learning Lab was an excellent opportunity to talk about GPT-3, a recently developed artificial intelligence model capable of realistic text generation (e.g., news articles or chatbots that appear human). During this session, our team showed the evolution of AI over the last decade in regards to interactions with robots and demonstrated the limit of each presented technology. Our team concluded the presentation with an amazing demo on how GPT-3 can generate fake tweets and gave some tips to the audience on how to potentially detect fake content.
In order to address another “hot topic,” which was stated as the n°1 “fear” by the general public during last year’s “TELL” session at Digital Days, Unit8 partnered with Kudelski and the The Swiss Academy of Engineering Sciences (SATW) to organize a debate on the use and security of personal data. During this session, questions about data security and trust were answered by a panel of experts from Kudelski Security, Vaud’s Directorate-General for Digital and Information Systems, QoQa, SBB/CFF, and of course a Data Scientist from Unit8, Rudolf Höhn.
Source: TELL, the digital pulse of the population, Olivier Wyman, 2019
Despite the amazing evolution of AI when it comes to interaction between human beings and “robots” (from Cleverbot in 2006 to more recent examples like Siri or Alexa), these technologies still show some limitations. With the arrival of GPT-3, today’s most advanced text-generation model, it is easy to assume that technology may have surpassed a new milestone. However, the use of GPT-3 is still experimental at this stage, which makes right now the perfect time to sensitise the general public on how to detect fake content (for example, by tracking inconsistencies in texts or, in the case of chatbots, testing answers with unrealistic questions).
In the long term, we will have to face the issue of the responsibility of AI-based systems in the case of fake news. The issue is not whether the technology is good or bad, but rather whether the responsibility for the use of the technology is with the creator or with the user. The answer to this question will be crucial for the future of AI.
The last paragraph was not written by us, but copied and pasted verbatim from a GPT-3 private platform after giving it the first written sentences — that’s how good the text generation is. When we tested this type of content on humans, fake content fooled even the experts working at Unit8. In a limited internal test, over 50% of our data experts voted that fake tweets were real!
So, what comes next? Will this technology completely change our lives? We don’t know, but GPT-3 will have no trouble at making a guess.
The first part of the debate clarified what types of data private and public entities use and how they store this data. Unit8 outlined that data does not always mean “personal data” and explained how AI can be also used to help the industry solve business and operational challenges or optimize internal processes.
One can observe some rising concerns and lack of trust about not only GAFAM but also other private and public institutions. In order to tackle this issue, the speakers agreed that:
The debate ended with the notion of “digital hygiene”: the responsibility of the customer to choose the companies they agree to give their data to (ethical companies) and the necessity to regularly investigate and ask questions about where data is stored and what data should be kept online.
In its fourth year, Digital Switzerland was organised in collaboration with more than 100 partners. More than 80’000 people attended these 3 days of online and physical sessions. The most popular topics and hence the main concerns of the Swiss population were transparency, security, and the impact of digitalisation on our future and personal freedom. 200 Learning Labs were organised and 3’500 people were able to improve their digital skills. Unit8 is proud of having participated in this impactful event and for having contributed to demystifying AI and answering some questions from the Swiss population about their digital future.