Fullscreen HTML5Med Res (??MB)HTML4

One step closer to a dystopian abyss? Techno Talk Max the Magnificent As always, we live in exciting times. Indeed, times are getting more exciting by the minute. Can you imagine being able to simply look at something, ask a question, and receive a spoken answer from an AI? I just saw such a system in action! I n my previous column (PE, April 2024), we talked about the concept of mixed reality (MR), which encompasses augmented reality (AR), diminished reality (DR), virtual reality (VR) and augmented virtuality (AV). Mixed reality is exciting, but the real game-changer will come when we combine it with artificial intelligence (AI), all boosted by the awesome data bandwidths promised by mmWave 5G and 6G cellular communications. The question is, whether this will be a game-changer for good… or the other sort. Where are we? A large language model (LLM) is an AI model notable for its ability to achieve general-purpose language understanding and generation. The first LLM to impinge on the general public’s collective consciousness was ChatGPT. Created by OpenAI, ChatGPT began to roam wild and free in November 2022, which is only around 18 months ago as I pen these words. This form of Generative AI (GenAI) is now all around us. There are AI-based writing tools (give them a few text prompts and they will write your marketing slogans, product descriptions, brochures… and so on); AI-based presentation-generation tools (give them a few text prompts and they will generate your PowerPoint presentation for you); AI-based speech-to-text transcribers (give them an audio or video file and they will return the written transcript); AI-based content summarisers (give them an audio or video file – or the output from a transcriber – and they will return a summary along with a list of action items); text-to-image generators (I’m currently having a lot of fun with Stable Fusion), and – most recently – a company called DeepMotion announced a text-to-3D-animation tool called SayMotion. In my case, I’m particularly interested in how AI might help with hardware design and software development. As a case in point, shortly before I started to write this column, I whipped up a tiny test program to run on an Arduino Uno. 8 This comprised only 34 lines of code, 20 of which were items like { and }. Out of the 14 lines containing more meaty statements, 11 of them (that’s close to 80%) had bugs, and this was one of my better days! Prior to the introduction of LLMbased assistants called copilots, embedded software developers typically spent 20% of their time thinking about the code they were poised to write, 30% of their time writing the code they’d just thought about, and 50% of their time debugging the code they’d just written. By comparison, 60% of today’s embedded code is automatically generated by GitHub Copilot. This would offer a tremendous performance boost if not for the fact that – since Copilot was trained on opensource sources – 40% of the code it generates has bugs or security vulnerabilities. Fortunately, we have Metabob from Metabob (don’t ask), which is a form of copilot that identifies and addresses any problems introduced by humans and other AIs. Where are we heading? There’s a famous quote: ‘It is difficult to make predictions, especially about the future.’ This quote is so famous that no one knows who said it. It’s been attributed to all sorts of people, from Mark Twain to Niels Bohr to Yogi Berra. Whoever did say this knew what they were talking about. I would never have predicted many of the technologies we enjoy today. Contrariwise, some of the technologies I was looking forward to seeing have failed to materialise (in more ways than one). One of the questions I often ask technologists when I’m interviewing them is, ‘Will we have technology XYZ next year?’ (where XYZ is whatever futuristic technology forms the topic of our conversation). Of course, they always answer ‘No.’ My next question is, ‘Will we have this technology in 100 years’ time?’ To this, they always respond ‘Yes.’ Then I say: ‘So, now we have the endpoints, all we need to do is narrow things down a little.’ I am confident it won’t be long before we are all sporting some form of MR ‘something or other.’ Personally, I think one of the MR interfaces intended for daily usage that will arrive on the scene sooner rather than later will look a bit like a pair of ski goggles, but I’m prepared to be surprised by something else. When people tell me that they would have little use for an AI+MR solution, I think back to the early 2000s when the same folks told me they had no use for phones that could take pictures (‘All I want to do with my phone is make calls’). All I can say is, ‘Look at you now!’ One of the examples I often present is being able to ask my AI+MR combo a question like, ‘What was that book I was reading a few months ago that talked about AI and Ada Lovelace?’ I can envisage the AI responding with the name of the book, while the MR highlights its location on my bookshelf. There are several required ‘building blocks’ that are starting to fall into place. For example, a company called Prophesee makes a teeny tiny eventbased vision sensor that’s only 3mm x 4mm in size and consumes only 3mW of power. Another company called Zinn Labs has mounted these sensors in a pair of glasses frames that are also equipped with an eight-megapixel forward-looking camera, microphones, loudspeakers, and a cellular connection to a cloud-based AI. The camera captures the scene while the sensors track what your eyes are looking at. I’ve seen a demo where the user looks at something like a plant and simply asks a question like, ‘Can I grow this plant indoors?’ The AI immediately responds, ‘Yes, but it requires bright light, so you’ll need to place it near a south-facing window.’ Horns and swords I feel like we are sitting on the horns of a dilemma with a Sword of Damocles hanging over our heads (I never metaphor I didn’t like). I hope we’re heading toward an age of wonder; I fear we’re one step closer to a dystopian abyss. Pass me my dried frog pills. Practical Electronics | May | 2024