GlobalSaké 2021 panel on AI/ML
Last Thursday, GlobalSaké 2021 organized another panel session under their PalamINT Annual Series 2021. The title of the panel was “Multilingual Artificial Intelligence, Natural Language Processing (NLP) and Applied Machine Learning”. To discuss the topics, 4 highly skilled professionals in the field were invited: Jane Nemcova (MIS), Omar Siddiqui (LinkedIn), Rafal Jaworski (XTM International), and Jorge Penalva (Lang.ai). The discussion was moderated by Dave Ruane (XTM International).
Besides the strong set of panelists, another component that made the event special was its audience. This one was formed by a mix of personas surrounding AI and Language Technologies, including linguists, engineers, salespeople, product managers, and participants curious to know more about the topic.
The interesting discussion among the panelists, first stimulated by the moderator and later by the audience’s questions, brought some interesting takeaways about AI and Natural Language Processing. I will explain them in further lines.
What is your factual knowledge about AI and your relation with the field?
Before the discussion started, GlobalSaké 2021 organizers conducted a very entertaining quiz, measuring the general knowledge of the public about AI/NLP. The first question was about who was the first to coin the term “Artificial Intelligence”: Noam Chomsky, Elon Musk, Alan Turing or John Mccarthy. Another question concerns which was the first successful NLP application accepted by general consumers: Alexa, Google Assistant, Siri, or Microsoft’s Cortana. Do you have a guess? You are correct if you choose John Mccarthy and Siri, respectively!
In this first part of the event, the organizers also stimulated the audience to tell a bit more about their experience with the topic. They were asked what is the first word that comes to their minds when they hear AI. Participants pointed to related terms such as “robots”, “machine” and, funnily, “Will Smith”! Next, organizers invited the audience to tell some areas in their lives where AI/NLP is used. In exchange, the public mentioned work, machine translation, Netflix recommendations, and famous conversation assistants, as you may see in the picture below.
After the quiz, the GlobalSaké 2021 panel starts with the moderator asking why Natural Language Processing is so relevant nowadays. For the panelists, such relevance comes from the easier access we have nowadays to language technologies. This statement may be confirmed by the big number of applications mentioned by the audience for one of the questions in the previous quiz and depicted in the figure.
Moreover, another relevant point mentioned by the panelists is the news coverage that has been given to the possible risks of state-of-the-art technology in Natural Language Processing, known as Large Language Models. Such concerns mainly rise from a scientific article entitled “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?”, which points to some of the negative implications such technologies may have.
The discussion then moved on to a fundamental question: “What is Natural Language Processing?”. Many of the technical terms involved in the scientific literature were mentioned at this moment, such as word embeddings, neural networks, context, BERT, GPT-3, fine-tuning, and so on. Moreover, some of the limitations from the current state-of-the-art were discussed.
First, the panelists mentioned that the big majority of NLP applications nowadays are focused on the English language, suggesting we should fix this bias by developing diverse models in terms of the languages they process. Moreover, they pointed to the fact that current state-of-the-art technology in NLP (e.g., large language model) works as “black-boxes”. In other words, although such language models can solve a large number of tasks with high accuracy, we still do not understand perfectly the sequence of steps they took to get into the final solution.
If this phenomenon is already non-optimal when the model makes the right decisions, it can be even worse for the wrong one. As a solution, if the model generates an unpleasant output, understanding the sequence of steps it took may help to explain the final bad decision as well as fix it. That is why working to improve the transparency and explainability of these models is so important.
Questions of the Audience
In the end, GlobalSaké 2021 organizers opened the room to the audience so they could engage in the discussion by doing questions. The first one asks the panelists which would be their guesses regarding the next financial-wise successful NLP use case. One of the panelists reminded us we are living in a time where diversity has been celebrated.
In this context, large corporations have looked for ways to mitigate strong bias in all their fronts, including the language used in their platforms. According to the panelist, NLP solutions that help to automatically detect and fix linguistic bias may be the next financial-wise success use case in the field.
Another interesting question concerned the working positions in the field of AI. Specifically, a participant of the event asks about which role a professional in Linguistics could assume to work with AI/NLP. Panelists argued about the unique period we live in and how things change more frequently and fast. In such times, they stated that companies need to work in optimal solutions to match (traditional or novel) skills of work candidates to the company’s needs.
On the other hand, workers may carefully look into the job requirements and argue to their potential employer how their skills can be useful to the target task or job. Finally, to learn a new skill, panelists also remembered the wide range of easily accessible courses that exist nowadays in MOOC platforms such as Coursera and Udacity.
In conclusion, GlobalSaké 2021’s session on “Multilingual Artificial Intelligence, Natural Language Processing (NLP) and Applied Machine Learning” was a success thanks to the strong set of panelists and an audience with a diverse background. It is interesting to see how the topic has raised the attention of many people who are interested and looking forward to being involved with such technologies.
By attracting such a diverse group, we understand the importance of these technologies being diverse by themselves. And to achieve this goal, from the discussion with the panelists we could conclude that working on the transparency and explainability of these technologies is an important thing.