Crazy singularities, robot rebellions, falling in love with computers: Artificial intelligence conjures up a multitude of wild what-ifs. But in the real world, A.I. involves machine learning, deep learning, and many other programmable capabilities that we’re just beginning to explore. Let’s put the fantasy stuff on hold (at least for now) and talk about real-world A.I. Here’s what it is, how it works, and where it’s going.
A.I. seeks to process and respond to data much like a human would. That may seem overly broad, but it needs to be: Developers are baking in human-like smarts into a wide variety of applications. Generally, A.I. falls within three categories — which we would note there is still some disagreement as to what the exact definitions are, much less if they’re truly possible.
A.I. can also be classified by how it operates, which is particularly important when considering how complex an A.I. system is and its ultimate cost. If a company is creating an A.I. solution, the first question must be, “Will it learn through training or inference?”
As we’ve noted earlier, these definitions are only meant as a general guide (this Medium article is a great discussion on what we’ve just talked about), and some may have slightly different descriptions. But there are examples of current A.I. which are worth discussing.
Voice assistants: Siri, Cortana, Alexa, and other voice assistants are growing more common, becoming the “face” of modern A.I. A growing subset here are chatbots, which manage messaging on websites and carry on online conversations.
Translation: This isn’t just about translating language. It’s also about translating objects, pictures, and sounds into data that can then be used in various algorithms.
Predictive systems: These A.I.s look at statistical data and form valuable conclusions for governments, investors, doctors, meteorologists, and nearly every other field where statistics and event prediction prove valuable.
Marketing: These A.I.s analyze buyers and their behavior, then choose tactics, products, and deals that best fit said behavior. There is a lot of crossover between these behind-the-scenes tools and voice assistants at the moment.
Research: Research A.I.s like Iris search through complex documents and studies for specific information, typically at higher speeds than Google’s search engine.
Awareness: These A.I.s watch for and report unusual events when humans can’t have an eye on them. One of the most complex examples of this is theft detection, which reports unusual behavior. A more exciting example, however, is self-driving cars, which use A.I. systems to scan for dangers and choose the appropriate course of action.
Editing software: These basic A.I.s look at pictures or text and locate ways that they could be improved.
Recently, neural networking expert Charles J. Simon recently opined on our pages about where he thinks A.I. is headed, which we recommend you read. While we won’t cut and paste the entire article here, we’ll point you to one specific section:
Most people look at the limitations of today’s A.I. systems as evidence that AGI [general A.I.] is a long way off. We beg to differ. A.I. has most of AGI’s needed pieces already in play, they just don’t work together very well — yet.
This is a key point. As we’ve noted, A.I. is getting better — at least perceptually — by the fact that developers are stringing together several narrow A.I. platforms. But the platforms don’t talk with each other. For example, while Alexa might now be able to start your car, it can’t use the current weather conditions to adjust your car’s heater or air conditioning systems or start the defroster to make sure you’re ready to go as soon as you get in. But Simon argues that we may have the computational and developmental capability either already and don’t know it yet, or within the next decade.
Companies are spending massive amounts on money on A.I. right now, and as long as they’re willing to spend the billions (if not eventually trillions) to advance the technology, things are going to move quickly. But there are all kinds of roadblocks in the way — whether it be a recessionary economy, computational challenges, and even moral and philosophical hurdles to overcome — so the road to a real-world Skynet might be a long one.
While we keep coming back to the obvious Skynet references, it’s time for a bit of a reality check. A.I.s are long strings of programmed responses and collections of data right now, and they don’t have the ability to makes truly independent decisions. That being the case, malice is definitely off the table for the time being. But that’s not to say human error could make them so.
For example, if a predictive A.I. tells a team that storms will spawn on the East Coast next week, the team can send resources and warnings there in preparation. But if storms actually appear in the Gulf of Mexico and hit the coast there, that prediction was inaccurate and may have endangered lives. No one would think the A.I. is somehow personally to blame for this; instead, they would look at the various data inputs and algorithm adjustments. Like other types of software, A.I.s remain complex tools for people to use.
At least for now, A.I. is, for the most part, harmless and if anything helpful to the world at large. But that could change in the distant future, and at that time we’ll need to have a serious discussion on just how much of our lives we’re willing to turn over to machines.
San Francisco startup Dimensional advertised for people to walk robot dogs through city streets. The New York Times reported that the job could include carrying a roughly 33-pound robot back to the office if it overheated. A lift home is an awkward requirement for a labor-saving machine.
Keeping the hardware running is only one challenge. Anthropic's robotics tests found that language models mostly failed when controlling joints directly, but did better when directing trained controllers. Giving an AI agent a body doesn't teach it to handle ordinary household objects.
Soon, a mattress protector may be more than a bed cover. Australian medtech company Sleeptite and researchers at RMIT University have received a US patent for FlexiSense, a smart bedding technology that tracks body position and posture to help caregivers identify when older residents may be at risk of falling.
The technology is being developed for Sleeptite’s in-bed monitoring system, REMi, and was recently on display at The MedTech Conference in Boston. A peer-reviewed study in Flexible and Printed Electronics says the sensors can also be made at scale without sacrificing performance or durability.
AI companies spend enormous amounts of time convincing us their technology is safe and secure. But what happens when those same companies become targets of increasingly sophisticated AI-powered attacks? Anthropic, the company behind Claude, is now facing that question. According to a Washington Post report, cybersecurity startup Cathedral discovered vulnerabilities that let its researchers access sensitive information in Anthropic's internal systems.
The researchers used AI agents to identify potential entry points in less than 12 hours. Interestingly, they didn't need to crack Claude itself. Instead, they exploited weaknesses in third-party business applications that Anthropic uses for everyday operations. Anthropic's security testing program disclosed the findings, which allows outside researchers to identify and report vulnerabilities.