For around a decade, the promise of artificial intelligence (AI) has loomed over us. It started inconspicuously, with games — first chess, then classic strategy games, and even Go, which was once thought to be purely beyond the reach of computers. But games are well-defined, deterministic, and limited. Real life is much different.
However, in the past year or two, AI stopped being a games-only technology and moved up into the real world, promising to shake things up in fields ranging from writing to biochemistry. Here are just a few of the big impacts AI has made very recently.
Generating text
We couldn’t start the list without ChatGPT, the text-generating AI that’s been making waves since late 2022. The AI is essentially a chatbot developed by OpenAI, an artificial intelligence research laboratory founded by Elon Musk and Sam Altman, among others.
ChatGPT was launched as a prototype on November 30, 2022 and since then, it took the world by storm. The technology itself isn’t that new — we wrote about GPT-2 and GPT-3, the precursors of ChatGPT. It’s the same approach, but no doubt, ChatGPT is just better.
The chatbot was so successful it’s already reached 200 million users, and large companies like Google are starting to see it as an existential threat to their existence — so much so that they rushed to release their own version of the chatbot, which flopped spectacularly and sent the company’s stocks spiraling down.
Surely, bots like ChatGPT have the potential to change the world, and this is no longer a prospect for the future, it’s happening now. But ChatGPT and its peers still have plenty of limitations. For starters, they suffer from algorithmic bias, and while its creators have attempted to limit any show of bias, users were able to make ChatGPT generate replies indicating that women and scientists of color were inferior.
But perhaps most significantly, ChatGPT has no understanding of what it’s actually saying. It’s a text predictor, and while it can be very convincing, it’s still just predicting text, not creating anything per se. This makes it very useful as a general chatbot, but also as a productivity tool (it can, for instance, write basic programming code or create the outline for a project), but it’s not that great as a creativity tool.
Still, it’s hard to gauge just how big its impact will be quite yet. In the couple of months since it showed up, ChatGPT has rocked the internet, and its peers are right around the corner, waiting to make their mark.
Generating images
They say an image is worth a thousand words, and surely, computers can’t generate realistic images. Or so we thought.
Images seemed way more off-limits than words, and yet, we’ve reached a point where it’s not just one or two, but several AIs that can produce convincing, realistic, and stylistically adaptable images. Don’t believe me? Just see for yourself.
Image-generating AIs are so good that it’s not just artists who are embracing them (an AI won an art competition, by the way). Architects and other professionals are also embracing this technology. Sometimes, it can go off the rails, but as people experiment more and more with elaborate inputs, as more data gets fed into the algorithms, the quality is improving more and more.
Yet again, technology has evolved quicker than policy, and artificially-generated images have taken the world by storm. For now, algorithms don’t allow sexual content, or content that would be insulting or detrimental to people’s image — but who will enforce that as time goes on, and how? The prospect is exciting, but the threats are also very real.
Audio filtering
Something that won’t make that much of a splash but is definitely going to be impactful for some industries is audio filtering. Whether it’s noisy interviews, bad production, or any other situation where the sound isn’t clear, having a way to filter out the unwanted sounds can be a godsend.
Sound processing is nothing new, but AI seems to be elevating things to the next level. For instance, Adobe released a free AI-powered audio processing tool that can enhance poor audio recordings and make them much better. We’ve actually tested it and while it won’t save every bad recording, it’s significantly better than any other tool we’ve tried.
Audio to text
As science communicators, we work a lot with interviews and have spent quite a few hours transcribing discussions with researchers. Up until a few years ago, this was done manually: you listen to the recording and you transcribe it. More recently, a number of services provided automated transcribing services, but most were expensive, bad, or both. But recently, AI started taking over.
Thanks to AI, online transcription has substantially improved its performance, and is nearing (or has already reached) the point where it can be fully automated. It’s not the biggest impact AI is making, but it’s definitely one that we’re thankful for.
… and text to audio
Transcribing audio is one thing, but producing audio from text is another. Of course, text-to-speech generators have also been around for some time, but what if I told you that you can recreate anyone’s voice and use it to say anything? It sounds amazing and creepy, right? Well, what if I told you the new Top Gun movie and Star Wars are already doing it?
But wait, it gets even better: music companies are already claiming that they can train AIs and get people’s voices to sing whatever song they want. It’s not just vocals, either: Google recently released MusicLM, an AI-based music generator that can turn text prompts into audio segments.
Decoding protein structure
If you’re still not convinced by the impact that AI can have, there’s one more argument we have: proteins. Proteins serve a wide range of purposes. Some transport molecules, others offer structure or serve as receptors, and so on. Each of these functions is closely related to its specific shape, which is achieved through folding. Predicting protein shape is important for biomedical science, but it’s notoriously difficult — which is why it’s so impressive that AlphaFold, another AI developed by DeepMind, managed to crack the structure of virtually all proteins known to science, hundreds of millions of them.
While there’s still plenty of room for improvement, this could usher in a new wave of discoveries in the biomedical world, accelerating almost every field of biomedical research.
The bottom line
There’s still plenty of room to improve things, and we’re still in the early days of artificial intelligence research. But AI is no longer a technology that holds a future promise, it’s something that’s making an impact now. Other fields, like self-driving cars or electricity distribution could also greatly benefit from AI in the not-too-distant future, and we may just be seeing the tip of the iceberg in terms of what AI can do in society.
Hopefully, we’ll manage it wisely.