The future is coming, but before then, we’re still a long way from the next version of a machine-vision tool.
The first version of AI tools like DeepMind’s DeepNet, which are built on deep learning algorithms, was released last year, and they’re still in beta testing.
But it’s hard to say how close we are to the full-fledged tool, as many of the AI software that we’ve seen recently is built on top of existing AI techniques.
“If you look at the early years of DeepNet [a tool used to build deep learning systems], you can see that the algorithms have matured, and now you see that we have a lot of work to do in the pipeline,” said DeepMind CEO Demis Hassabis.
“And that means that there are lots of new things to build on top, and so I think we’re going to get there very quickly.”
But it also means that we won’t see the next generation of AI software in time to compete with the likes of Google’s machine-building capabilities, which have been around for decades.
That’s why Hassabis thinks it’s important for AI researchers to get their hands on tools like deep learning, as they can help to speed up the development of AI algorithms, and give them the ability to build more powerful and efficient models.
“The next generation is really going to be really exciting,” he said.
“It’s going to require a lot more work.”
But even if DeepMind eventually manages to release a new version of its AI system, it’s not a guarantee that we’ll see it soon.
“There are a lot tools out there now that are based on deep neural networks that have already been used in AI, and if you look in the literature, you’ll find lots of people who were using those in the early days,” Hassabis said.
It’s also not clear how far along deep learning will be in its development.
“A lot of these [deep learning] algorithms have been in the wild for decades,” said John H. Tetzlaff, a professor of computer science at the University of Southern California and a co-author of a recent paper on the state of deep learning.
“We don’t have enough data to say that it’s mature enough to compete.
But in a lot a different ways than you might think.”
If we don’t get any of the deep learning tools that we want to see in the next decade, the world will look a lot different, he said, adding that we’re at a “critical point” where we’re starting to see “a new wave of AI.”
So while the future of deep-learning AI is still a ways off, there are some things that we can expect to see come out of the process in the coming years.
“I would say that there will be some AI tools coming out that we haven’t seen before,” Hassabas said.
There’s also a chance that some of the new tools we’re excited about will just come from other research groups, like the ones he mentioned.
That would be a welcome development for AI research, which is focused on developing deep-entering tools that can help scientists and engineers build machines that can learn and think and act in ways that are useful to us.
“But if you’re talking about something like machine-to-machine, that’s going be different,” Hassabi added.
“For example, what we’ve done with deep learning for example is have trained a very basic model that’s just a bunch of data.
And then you have the machine learning algorithms that have been built on that model.
We want to build a model that can actually learn something about the world, but also be able to act on the world.
So it’s the combination of those two things that makes it so much different from anything we’ve built before.”
But that’s not to say we won-‘t see some of these tools come out, too.
“That’s where we get to where we need to be,” Hassabee said.
In the meantime, there’s one thing that we should definitely not expect from these new AI tools, however.
“These are just the tools that are out there and we’re not sure how good they will be,” Tetzliaf said.