[ad_1]
In today's pop culture, machines with artificial general intelligence (AGI) are generally described as analogous characters, speaking and speaking, brimming with personalities – from the deadly intent of the Terminator to the Nobel heroism of Vision. In reality, conscious robots are far away. Nathan Michael, Research Associate Professor and Director of Carnegie Mellon University's Laboratory of Resilient Intelligent Systems, says widespread AI systems will develop from "narrow" IA systems. unique current.
"The general artificial intelligence is representative of this concept of bringing together many types of specialized AI," he explained. Michael explains that "AGI is not so much an autonomous system – it's not a digital Athena emerging from the Zeus front – but rather a threshold of capacity derived from a whole set of people." Close AI working together.
Michael compares him to a baby. When a person is born, she does not have a conscience or a sense of self. There is no overall mental operating system in place, motivating their actions. "They develop different forms of specialized organic intelligence," he explained. These specific shapes allow the person to observe his environment, to differentiate objects, to pick them up and to move around. "It's the combination of a specialized artificial intelligence that creates an increasingly sophisticated specialized intelligence that allows this organic smart system to become more and more capable," he concluded.
The same essential process occurs in the development of AI. You can see it in Flippy's Miso Robotics. Originally, this chef robot was only able to do one thing: return hamburger patties to a grill. It has since been upgraded to operate a deep fryer and the company is teaching it the ability to clean itself. With enough time and enough upgrades, Flippy could become what we consider a "general" AI in that it would be able to do everything that a human would be able to do in the future. cooked.
Getting to this point, however, is fraught with pitfalls. Technically, processing power, as well as storage and data management, are limiting factors of the artificial intelligence capabilities that we can currently achieve. Michael points out that, thanks to advances in data storage, "we are starting to ask ourselves questions: how do we understand the nature of the data we have, how does this affect algorithm performance and performance? affect the overall performance of the system. "
"And that brings us back to that earlier challenge of simply understanding the accuracy of the algorithms themselves to integrate them into different contexts or conditions," he added. Michael cites the many cases of bias inherent in training data. "If we understand the nature of the algorithm, we understand how it behaves in the face of different types of data inputs and how that variation will impact the performance of the algorithm."
As we start to brew more AIs, the complexity of measuring their combined performance increases exponentially. "So, be able to talk about the performance of these algorithms and understand this more deeply," Michael said. "We can understand how these systems can be combined significantly."
This need to understand what is happening in the spirit of our mechanized creations comes from our need for trust. "Confidence in the algorithm, trust that it works as expected, trust in the understanding of how the performance of this algorithm changes based on the data," Michael said.
"Historically, as we tinker with technical systems, we are talking about establishing a relationship of trust over time, with real evidence," he continued. "And this type of model will of course be transferred here because, although we sometimes speak of AI as nebula, it is in fact only a technical system." Just as we expect Siri today to provide us with accurate weather reports, and our GPS systems are keeping us out of the cliffs, tomorrow's RNs in general will need to gain the trust of their users to be widely adopted.
Bringing people on board with these more and more advanced AI systems will probably not be too difficult. Of course, we have a stomach ache when robots come to occupy all of our most junior jobs, the most dangerous and the lowest paid, and collapse with T2 Memes every time Boston Dynamics launches a new robot, but humans have shown themselves more than ready and willing to adapt their behavior to new technologies.
"We live with AI," Michael said. "Every day, it becomes more and more sophisticated, in terms of individual AI capabilities or the combination of these capabilities, and so I think we are seeing this human adaptation now, day-to-day. "
Michael goes on to explain that, while many advances in AI will be rather mundane (think that Siri is gradually becoming more proficient), some changes will have immediate and major repercussions on society.
"There will be moments of transformation, for example when we will be able to believe that autonomous vehicles can move us from point A to point B," he said. "They no longer need an individual driver's license because they no longer drive." Michael also explains that these kinds of "moments of transformation" occur fairly regularly during the course of history. He cites the example of the car without horses.
"What we saw there, is that there was a short period of time during which these machines were present on the same roads as horses and people," he said. declared. "There was not a well-established understanding of how these different types of vehicles and people should be engaged." However, the men of the time (if not the horses) were able to gradually adapt to the presence of the new technology, partly to its gradual introduction. "We are seeing small transitions over time in the way society interacts with the technology itself, and that's certainly what we're seeing now," concluded Michael.
Indeed, just look at Google's efforts over the past year since the formal establishment of its AI division to uncover evidence of increasing infiltration of AI into our business. everyday life. Take Duplex, for example. This AI is designed to make restaurant reservations over the phone by listening to and analyzing human language and then responding accordingly. It's an incredibly complex business that requires countless hours of R & D. In less than 12 months, we have not only seen the service extend to the smartphone ecosystem and to 43 US states, but also the introduction of a complementary artificial intelligence service for companies, also called CallJoy.
In addition, although the reaction to this advancement was initially a source of concern, this feature has become commonplace – an additional feature of Google Assistant. And with the company's latest breakthrough in the field of machine learning, the wizard will soon be bothered by network connectivity, an advance that will integrate all the power of Google's artificial intelligence. in all smartphones.
[ad_2]
Source link