AI Faces EU Regulation, 6.9 million people get their DNA data hacked & Google fakes a demo | The Hi-Tech Discussion

After OpenAI's massive near-meltdown and Elon's exciting week of flipping the proverbial bird to Bob Iger and all advertisers, the Tech industry seems to be back in its normal flow of news and events.

Nonetheless, this past week hasn't lacked interesting developments, from the EU agreeing on AI regulation to Google coming up with an exciting (and fake) Gemini showcase, and a DNA company getting hacked.

Welcome back to the Hidalgo Tech Discussion, the weekly series that saves you hours of news reading and puts everything under one article.

23andMe: Hackers steal the data of 6.9 million users

Full transparency: I've done DNA tests – fortunately, not the 23andMe one but the one from MyHeritage . The last thing you want and expect is for one of these companies to fail in storing your genetic information safely.

You could tell me "Well, Mr. Hidalgo, you're the dumb one. Why would anyone give their DNA to a company?"

I'm curious, okay? And finding out that I have 11 ethnicities was quite interesting, so for those who come at me with 'curiosity killed the cat' I reply 'but satisfaction brought it back to life.'

Now back to our topic.

On December 1st, 23andMe announced that hackers gained access to 0.1% percent of their user base's data. About 14.000 users.. with a catch: The company actually meant 6.9 million. Of course, a small and totally unintentional calculation error, right?

Now, let's be clear: This is unforgivable, and the company might still be lying, considering that they didn't seem to confirm or deny the reports users in hacker forums selling the data of a total 5 million individuals two months ago. Some of that data was openly advertise as belonging to people (like myself, even though in just 5%), of Ashkenazi Jewish descent).

But 23andMe went a step further, and they blamed the issue on customers re-using common passwords... lol. Imagine someone breaks into your house, you call the police and they refuse to help or investigate because you didn't invest in stronger doors.

It's one thing if someone hacks a million Instagram accounts and finds the chats filled with silly cat videos you send t your friends, but it's entirely different when the information they find is your genetic makeup and your close relatives.

Lives are at risk here. And if regulators and the justice system don't make an example out of 23andMe, things will only get worse.

Here's my suggestion to 23andMe: Hire somebody to help you shape a new narrative. You need to show compassion, understanding and support to your user base, and you will probably need to come up with some kind of compensation. The main thing to salvage here is your reputation, but you do so only by siding with your customers.

And now, the topic of data brings us straight to the next section:

EU regulation is in... bad news, or long time coming?

December 1st wasn't eventful just for 23andMe: The EU lawmakers had to burn the midnight oil to finally strike a deal stemming from a 2021 proposal and months of discussions.

The result? The foundation of an EU-wide AI law.

Now, I am often critical of the EU and its regulations. They sometimes have a holier-than-thou attitude that comes across as quite hypocritical. Like for example banning some of the most populars tattoo inks due to "concerns for the health of people," while letting cigarettes and alcohol run completely free, despite the former being the main cause of lung cancer and the latter being the main cause of juvenile death (by car accident) and cirrhosis (a disease that puts a 5 year countdown on your head).

If you add to that the European mindset of fearing every new technological advancement or even actively opposing it as Italy did with Artificial Intelligence, you wouldn't be surprised to know I am greatly concerned every time EU regulation hits anything.

Now, there aren't specific details of the agreement, so all we have is a press release banning the use of AI in:

• Biometric categorisation systems that use sensitive characteristics (e.g. political, religious, philosophical beliefs, sexual orientation, race).
• Untargeted scraping of facial images from the internet or CCTV footage to create facial recognition databases.
• Emotion recognition in the workplace and educational institutions;
• Social scoring based on social behaviour or personal characteristics.
• AI systems that manipulate human behaviour to circumvent their free will;
• AI used to exploit the vulnerabilities of people (due to their age, disability, social or economic situation).

The only exception is law enforcement: Negotiators agreed that biometric identification systems can be used in public spaces for safety purposes with judicial authorization, for specific crimes, and for the targeted search of someone suspected of having committed a crime.

Now, you know business is my big passion, right? So, how does this regulation affect business?

SMEs will be able to develop AI solutions without pressure from industry giants controlling the value chain. Quoting the press release: "the agreement promotes so-called regulatory sandboxes and real-world-testing, established by national authorities to develop and train innovative AI before placement on the market."

To my surprise and against my never-ending skepticism, the EU came up with a first framework that sounds more than reasonable. Of course, there will always be concerned about how easy or difficult it will be for specific players to bend these rules, but that's a different conversation.

My final verdict on the topic: The EU gets a 9/10 for this. Congrats to Ursula von der Leyen and the European Parliament for delivering on a very important and difficult to achieve commitment.

Google's Gemini AI model debuts, but not without controversy

Google launched Gemini Pro this week as a tease for the alleged real deal, Gemini Ultra, which we know will arrive next year.

Let's start by saying that Google didn't get the "less is more" memo, and they structured Gemini like a McDonald's menu selection: Three options for you to choose from, of which there is one people will choose more often:

• Gemini Ultra: The real deal, full-blown AI model.
• Gemini Pro: A smaller, less capable model. Let's call it Crystal Pepsi because I get the feeling nobody will need this.
• Gemini Nano: For mobile devices.

You can try Gemini Pro on Bard, Google's ChatGPT killer. Gemini Pro can do better summaries, writing and brainstorming, and it beats GPT 3.5 – so if you don't want to pay 20$/mo for GPT 4, this might be a happy medium.

If you're not excited it's because not many people are, to be honest. Gemini Pro had one year to step into the ring with ChatGPT, and it still lost by decision... and then there's Gemini Ultra.

Gemini Ultra is basically GPT 4 with an ability to understand nuance in the input you feed it, and with better grades in math – a feature which I bet will make many a student quite happy. Another interesting feature is that Ultra will be able to answer questions about audio, video and photos.

However, here comes the sour part: We don't know much more about Gemini. In particular, we lack background on where its training datasets come from since Google refused to answer questions about it... bringing up once again the issue of data and intellectual property.

That is a wise decision considering that OpenAI is getting sued for IP violations, but it doesn't do Google any favors when it comes to speculation, concerns and skepticism like what you're reading now in this article.

The controversy

Want to know the kicker? I watched the entire Gemini demo and was pretty excited until finding out from TechCrunch that the thing was faked.

The incriminated demo showed Gemini mixing language and visual understanding, demonstrating a remarkable capacity for flexibility and illusion of reasoning. You can read more details here, because I'll skip straight to the fun part:

The video wasn't real. Gemini does not fully do the things they advertise in the demo, and the whole thing was a cherry-picking party of prompts matching still images made with the purpose of... crafting and narrative.

And here's where we enter my field.

You see, marketing and business are mostly about narratives: Perception is reality, so if you manage to shape and shift the perception of your audience, you are actively shaping their beliefs about what is true and what isn't.

Don't believe me? Well, Apple did a masterful job at this during their October event, when they claimed to be leading the USB-C transition while blatantly re-writing the reality: They were forced into USB-C adoption to conform to industry standards.

You see, people have short memories. Especially when they are emotionally involved – this works for products, relationships, politics, and any human endeavor. So, when Apple claims to be leading the transition, most of the user base will believe it.

Google tried to do the same: To craft a narrative. To paint a dream reality before it happened... but they got exposed, mainly because of a tactical mistake: They let the pressure of competitors get to them.

OpenAI is so ahead that Google felt hard pressed to do something about it, so they put together the launch of a half-baked version of ChatGPT, almost bringing us back to when Meta released Threads for it to massively flop. Then Google gave it a fancy set of names, tried to differentiate by making it complex, and then proceeded to lie about capabilities the tool doesn't yet have.

Now, better strategist would have been just releasing the demo of Gemini Ultra and then say nothing else about it. Or, even better, develop it to almost perfection and launch something that had at least a chance of blowing GPT out of the water.

Instead, they chose to lie to their customers. And you can re-write perceptions, but if you create a narrative so weak it has visible cracks, your house of cards will fall and you will be left with public distrust.

____________________________________

?? And that's it for this week.

?? Share this article with your audience if you found it useful, and follow me Sebastian D. P. Hidalgo for more weekly tech updates.