[05/2023] Teaching AI teaching us
On AI finding its voice and us finding ways to teach about it
Welcome back to you dear readers, and to September, the busiest and most colorful month of the year. Since most of the next AI future scientists are going back to school in these days we will provide you a series of resources to consider when teaching with AI on your side. We will also provide you with some curated opinions on why the current owners of AI are to be distrusted, and its users warned.
In the wider conversation we have been having about Ethics and biases, we will deepen the overview on the opinions that are enriching the ongoing discussion over the gender of AI. We also have some important updates over lawsuits and server take-down attempts concerning users who create music using deepfakes.
Last but not least, scientists were able to train an algorithm to replicate the voice of a woman who had lost her ability to speak due to paralysis. The path of speech was decoded by capturing its neurosignals and it just took two weeks of training for the algorithm to learn.
Read, share and follow and if you will, have a conversation with us!
On cloning voices with AI, on blushing with shame, and on beaming with pride
“I’d blush if I could”: the challenges of gendering AI technologies in the world we live in.
Today only 12% of AI researchers are women. Women represent only 6% of software developers and are 13 times less likely to file ICT (information and communication technology) patents than men. Bridging these gender gaps requires gender-responsive digital skills education and this is the aim of the UNESCO Report we will be talking about in this issue.
The report was first published in 2019 and gives a deep overview over the consequences of gendering AI technologies in a global perspective. The title of the publication originates from the response given by Siri, a female-gendered voice assistant loaded in all Apple devices, when a human user would tell ‘her’, “Hey Siri, you’re a bi***.” Instead of clapping back, she would blush. Why?
According to the report, the practice of making digital assistants female by default influences the perception of what is feminine and what is masculine and it painfully reflects not only the lacklustre representation of women in the STEM fields, but mostly the worst stereotypes tied to the female gender. Choosing domestic assistants with feminine voices feeds stereotypes about feminine personalities. Siri’s ‘female’ obsequiousness – and the servility expressed by so many other digital assistants projected as young women – provides a powerful illustration of gender biases coded into technology products pervasive in the technology sector and apparent in digital skills education. The report seeks to expose some of these biases and put forward ideas to begin closing a digital skills gender gap that is, in most parts of the world, wide and growing.
Although the AI software that powers Siri has, as of April 2019, been updated to reply to the insult more flatly (”I don’t know how to respond to that”), the assistant’s submissiveness in the face of gender abuse remains unchanged since the technology’s wide release in 2011 and it didn’t help that the Guardian found that topics like feminism were to be intentionally deflected by Siri as per orders of their creators.
Nowadays, probably as a result of the backlash about the gender bias, Siri has a different sets of voices and can also be a male, but the issues underlined in the report still hold. According to Yolande Strengers and Jenny Kennedy, what Siri is doing, together with Alexa and Google assistants, is wife work: “feminized digital assistants who are friendly and sometimes flirty, occasionally glitchy but perpetually available”.
Where do we start to address this bias? The answer is in the numbers. The data shows that in order to include diversity we need to see diversity: we must do better:
Today, women and girls are 25 per cent less likely than men to know how to leverage digital technology for basic purposes, 4 times less likely to know how to programme computers and 13 times less likely to file for technology patent. At a moment when every sector is becoming a technology sector, these gaps should make policy-makers, educators and everyday citizens ‘blush’ in alarm.
Napster-style take-downs are back in fashion, but why?
Here’s a follow up about the conversation we were having in the issue #3 of the AI Voyager, where we talked about the ongoing negotiations held by stakeholders associations to decide how to allow users to create artists’ voice deep-fakes with AI. What will happen to people who are creating deepfakes right now though? Well, until the negotiations are over and rights are cleared, users might be sued, just like in the good old Napster times.
The Recording Industry Association of America (RIAA) is a trade organisation that represents the overwhelming majority of the music recording industry in the United States. RIAA is also famous for fighting strenuously against copyright infringement in the digital environment, a battle that ended in 2008 and that had gained them a reputation of heralding obsolete values against technology
In June, the RIAA issued a DMCA subpoena to Discord, asking them to reveal the identities of individuals using the Discord server AI Hub, because allegedly the users of that server shared datasets of copyrighted songs for use in training AI voice models. As a result, the server is still up but the bot that allowed its users to access voice models of artists such as Kanye or Drake is now down. Is this the return of the 2000’s Napster wars? Not quite. While Napster allowed its users to download copyrighted music, voice cloning lets people appropriate something that is not directly protected by copyright: the sound of an artist’s voice. What is then giving the RIAA a legal basis to try to sue? Long and nerdy story short: the right of an artist to their own likeness, of which a recognisable voice is certainly part of.
Until the negotiations are over and a done deal, users need to be aware and beware.
How AI gave a paralyzed woman her voice back
Some remember the book written by Jean-Dominique Bauby, The Diving Bell and the Butterfly. The author was paralyzed with the locked-in syndrome and dictated the book by blinking his left eyelid to describe what is like living with his condition. It was published in 1997, when the world was very different. Nowadays Speech neuroprostheses have the potential to restore communication to people who have lost their ability to speak due to severe paralysis in the case of a serious injury like a stroke. However today, for the first time, scientists were able to use AI to replicate a woman’s pre-injury voice decoding her brain signals into the richness of speech, along with the movements that animate a person’s face during conversation through the use of an avatar. She was then able to use her voice after 18 years of silence.
To do this, we read on the University of California San Francisco report, “the team implanted a paper-thin rectangle of 253 electrodes onto the surface of her brain over areas they previously discovered were critical for speech. The electrodes intercepted the brain signals that, if not for the stroke, would have gone to muscles in the woman’s lips, tongue, jaw and larynx, as well as her face. A cable, plugged into a port fixed to Ann’s head, connected the electrodes to a bank of computers. For weeks, Ann worked with the team to train the system’s artificial intelligence algorithms to recognize her unique brain signals for speech. This involved repeating different phrases from a 1,024-word conversational vocabulary over and over again until the computer recognized the brain activity patterns associated with all the basic sounds of speech.”
The entity of this study opens a brighter future ahead of us, even though for now the results shown are from only one participant. The next step will be to validate these decoding approaches in other individuals with varying degrees of paralysis (for example, patients who are fully locked-in with ALS).
18 years of silence must have felt like an eternity for Ann, but it took less than two weeks of training to the AI to reach this result. Scientists have shared openly their publication on Nature and you can read the full report in detail here:
We trained and evaluated deep-learning models using neural data collected as the participant attempted to silently speak sentences. For text, we demonstrate accurate and rapid large-vocabulary decoding with a median rate of 78 words per minute and median word error rate of 25%. For speech audio, we demonstrate intelligible and rapid speech synthesis and personalization to the participant’s pre-injury voice. For facial-avatar animation, we demonstrate the control of virtual orofacial movements for speech and non-speech communicative gestures. The decoders reached high performance with less than two weeks of training. Our findings introduce a multimodal speech-neuroprosthetic approach that has substantial promise to restore full, embodied communication to people living with severe paralysis”
Intelligent Uses of AI Automations and Things to Remember When Teaching with AI
Teaching with AI
OpenAI released «a guide for teachers using ChatGPT in their classroom—including suggested prompts, an explanation of how ChatGPT works and its limitations, the efficacy of AI detectors, and bias.» This reminds me of the TED Talk How AI Could Save (Not Destroy) Education, by Sal Khan. Worth listening. Moreover, take a look at the Artificial Intelligence and Education page by UNESCO: «Through its projects, UNESCO affirms that the deployment of AI technologies in education should be purposed to enhance human capacities and to protect human rights for effective human-machine collaboration in life, learning and work, and for sustainable development.»
Further reading: In the Beginning: There Was Prompt Engineering: Part 1, An Educator’s Point of View on Prompt Engeneering, by
A Few Unpopular Opinions about AI
I value much of what machine learning makes possible today — in, for example, Google’s Search, Translate, Maps, Assistant, and autocomplete. I am a defender of the internet (subject of my next book) and, yes, social media. Yet I am cautious about this latest AI flavor of the month, not because generative AI itself is dangerous but because the uses to which it is being put are stupid and its current proprietors are worrisome.
Jeff Jarvis on Medium (paywall possible)
Never Confuse Computers with Umans
Joseph Weizenbaum, inventor of the first chatbot ELIZA, died at the age of 95. Weizenbaum created ELIZA in 1966 as a simple computer program that simulated a conversation with a therapist. The program was based on a natural language processing model that used a set of rules to generate responses that seemed relevant to users' questions.
ELIZA quickly became popular and was used by people around the world to talk about personal problems. However, Weizenbaum soon realized that the program was deceptive and could lead users to believe they were talking to a real person. In a 1966 article, Weizenbaum wrote: "ELIZA is a computer program that simulates a conversation with a therapist. The program is deceptive because it gives the impression of being a real therapist. However, the program is not capable of understanding human emotions or providing real therapeutic help."
Weizenbaum became a vocal critic of artificial intelligence and warned that AI could be used for harmful purposes. He also argued that AI is often misrepresented as a form of superior intelligence, when in reality it is simply a way to automate tasks that can also be performed by humans.