Hacker News Reader

Comment by nopinsight | Hacker News Reader

As a counter anecdote, one can also say that just 10 years ago, no computer systems came close to human at facial recognition or scene recognition for still images. Today's AI systems can caption events in videos. [1]

Researchers are also working on higher-level cognition tasks. We do not know for certain if there will be breakthroughs but there are many diverse tasks that seem within reach in the next 10-20 years.

What are human capabilities that you believe AI/Robots in 2045 will still likely be lacking?

[1] Dense-Captioning Events in Videos

"...We introduce the task of dense-captioning events, which involves both detecting and describing events in a video. We propose a new model that is able to identify all events in a single pass of the video while simultaneously describing the detected events with natural language.... " https://arxiv.org/abs/1705.00754

View on HN

There's a difference between technical capability in a laboratory/R&D context, and widespread adoption of a technology.

In 2045 we may very well have highly capable robots, but IMO there's not a chance they will be widespread to the point originally posited: that they will take care of (most of) the old, and produce (most of the) commodities.

One consistent theme when it comes to futurism - whether it's the Jules Verne vintage or the Popular Mechanics vintage is the consistent under-estimation of how much it costs to change the physical world.

Software will continue to be the largest source of change between now and 2045, because software is cheap and generally does not seek major edits to the physical world. Robots and other physical manifestations of futurism are very different.

The futurists of the past predicted domed cities, flying cars, orbital space resorts, mile-high towers connected by floating trains, robotic servants, and the such - none of it has come to pass, and the present world is dramatically less futuristic-looking than originally predicted, because major edits to the physical world is crazy expensive.

Software has been largely an end-run around this expense, by being (nearly) free to deploy to the masses. Deploying billions of household robots, or something similar, is far less likely to actually happen.

IMO a more realistic way to predict the future of 2045 is to imagine how much technological advancement you can cram into a world that still almost entirely looks like the present day. The physical world changes at a comparatively glacial pace.

While I broadly agree that things don't usually change how we expect, I do think the near future could invert that somewhat.

If we look at you examples of failed predictions: > domed cities, flying cars, orbital space resorts, mile-high towers connected by floating trains, robotic servants, and the such we notice that the real reason none of these exist is because they are all essentially impossible to achieve with our current technology.

However, robotic servants are partially already here. We can verbally ask a device to play a certain song or order us some loo roll, and it works. We have autonomous cars that can, reasonably safely, drive us wherever we want to go. Elon Musk's new factory is, reportedly, almost entirely automated. And as most of this progress is recent, it seems likely that it will continue to advance for a while.

I will agree that there is significant friction against changes in society, which slow down adoption of new tech, but consider that robots in the context we are discussing are not a consumer good, they are a replacement for employees. So perhaps the most relevant question that will drive or stall adoption is: Is a robot more cost-effective than a human.

I also think watching the extent to which self-driving cars are adopted by delivery/freight/taxi companies, and the fallout from that, will be a fascinating (maybe terrifying) glimpse into that future.

How is what you are saying different from someone saying 50 years ago that they can build both planes and cars, so the flying car is surely achievable in a few years?

The problem is that extrapolation is fundamentally hard.

As a counter counter anecdote, scientists where estimating that perfecting automatic text translation will take another 10-15 years... in ~1960. I guess hoping that future technology will solve all out problems is similar to expecting next generation to pay our debts.

> As a counter anecdote, one can also say that just 10 years ago, no computer systems came close to human at facial recognition or scene recognition for still images. Today's AI systems can caption events in videos. [1]

These "basic" computer vision tasks seem so trivial compared to the actual AI that the original AI researchers famously assigned solving them to a grad student over summer.

Reply on news.ycombinator.com