Looking back at how far AI has actually come since it all began

[SouthernBuffer]

The field formally traces its name back to a 1956 summer workshop at Dartmouth, where a small group of researchers coined the term artificial intelligence and genuinely believed a machine capable of human level reasoning might be achievable within a generation. That optimism collided with reality repeatedly over the following decades, producing what historians now call AI winters, stretches where funding and interest collapsed after early promises failed to materialize.

Expert systems in the 1980s brought a real commercial wave, rule based programs that could genuinely diagnose diseases or configure computer orders within a narrow domain, before that approach also hit a wall once the sheer complexity of encoding real world knowledge by hand became apparent. Progress then stayed genuinely slow and academic for most of the 1990s and 2000s, punctuated mainly by narrow wins like Deep Blue beating Kasparov at chess in 1997.

The actual turning point most researchers point to is 2012, when a neural network called AlexNet dramatically outperformed every other approach at an image recognition competition, proving that deep learning, an idea that had existed in some form for decades, finally had enough data and computing power behind it to genuinely work at scale. The 2017 transformer architecture paper then set up the specific technical foundation that every major language model since has actually built on.

ChatGPT's public launch in November 2022 is the moment this went from a research community's excitement to genuinely mainstream household awareness almost overnight. And the pace since then has been dramatically faster than almost any earlier period in the field's entire history, multimodal models, coding agents, and now systems that can complete genuinely multi step tasks with real autonomy.

So the honest arc here is nearly seventy years of alternating hype and disappointment before the current stretch of genuine sustained progress. Which is worth remembering both when today's pace feels overwhelming and when it feels like it might be slowing down again

[Mesh Ross]

Also, the transformer paper in 2017 barely made headlines outside the research community at the time.

Wild how much that one specific architecture ended up underpinning just a few years later

[TealBear]

The narrow AI wins along the way.

Chess, then Go, then image recognition, feel like a much clearer through line in retrospect than they probably felt at the time to people actually living through each individual moment

[Brooke_19]

Wondering whether people actually working in the field in the 1980s particularly saw the next winter coming. Or whether it felt as unstoppable back then as the current progress sometimes feels now

[Plateau70]

AI winters don't get discussed nearly enough in current conversations about the pace of progress. Funding and interest have collapsed hard at least twice before, no reason to assume the current momentum is permanently guaranteed to continue

[RogueDepot]

Deep Blue beating Kasparov always struck me as such a narrow kind of victory in hindsight.

A machine that could only play chess felt like a much smaller step than the framing at the time suggested

[Lantern98]

The 2012 AlexNet moment is really underrated as a turning point compared to how much attention ChatGPT gets. That image recognition result is really where the current era's technical foundation actually started

[Perigee Lewis]

Seventy years of alternating hype and disappointment is a quite sobering way to frame the current moment. Makes the current excitement feel less like an inevitable straight line and more like one particularly strong upswing in a much longer cycle

[Wardlow86]

The Dartmouth workshop researchers thought human level machine intelligence was maybe twenty years away back in 1956.

Worth sitting with how many times that specific prediction has been confidently made and missed since then

[FinnBalor99]

The expert systems era is a clearly interesting cautionary tale. So much confidence that hand coding enough rules would eventually add up to real intelligence, and it just never scaled the way anyone hoped