Gartner’s Technology Hype Cycle puts many application of artificial intelligence (AI) and Machine Learning (ML) in around the ‘peak of inflated expectations’ or headed into the ‘trough of disillusionment.’ Meanwhile, very few have made it onto the ‘slope of enlightenment’ let alone the ‘plateau of productivity.’
We’ve been working with AI for a while now and have seen real application and success in our own field. We believe that AI, when combined with other technologies, can be more than a game changer and can result in superpowers.
One reason that AI is in this trough of disillusionment is the height of its ridiculously high peak of inflated expectations. Science fiction has painted a picture of AI systems that are more powerful than human intelligence and capable of running complex ecosystems and even entire societies. The truth about AI is quite different. This truth is a reason to embrace rather than fear AI.
AI is at its best is when it is applied to tasks that exhibit certain traits. These are typically narrow in their application, but substantial in terms of the data utilized, particularly in the learning process. To put that into context, the Tesla fleet delivers driving data from every connected vehicle in use. This data is more than one million hours of driving, which is greater than a single person would experience in a lifetime. Another example is the creation of a supercomputer to beat the world masters as the board game ‘go’, which most people agree is the most difficult to play. The AI system could assimilate thousands of games and learn from them, resulting in an amazing level of ‘narrow’ intelligence, that could outplay even the most experienced human. This is entirely different to human intelligence, which is much broader and complete, something, contrary to sci-fi movies, is a long long way away.
Narrow, broad & general intelligence
So, narrow intelligence is when large sets of data are used to teach an AI system or a computer a simple task. The computer will seem extremely intelligent in its specialist subject, but dumb in just about everything else. This can, and has produced amazing results in tasks like early disease diagnosis or predicting trends in supply and demand.
Broad intelligence is less simple and requires the AI to perform a set of tasks, learning broader skills, but still within set parameters. An example of a broader system might be Amazon’s Alexa, or Apple’s Siri, with thousands of skills within limited parameters.
General intelligence is much more ‘human’ and currently completely out of reach for AI. This level of cognitive thinking is the stuff of future dystopias, run by computers who manage humanity with a rigorous set of objectives, resulting in decision that seem cold and heartless.
The application of AI in inspection
Inspection is a task that relies on viewing multiple images and analyzing the data to make decisions or to provide feedback to other parts of the manufacturing ecosystem. This is where AI can really thrive and deliver! It is a quick and relentless learner that can process data faster and more accurately than any human.
Take for example finished goods inspection of a medical pack. Once the package is sealed there is no way of seeing inside to ensure that every part is present and correct. In this instance, an x-ray image can be used to see the contents of the package and AI can be used to interrogate that image and ascertain if all parts or present and correct. The result is a fast, intelligent process that ensures that no partially complete packs leave production. This can be applied to just about any packaged product. This is a simple process of teaching the AI what a good pack looks like and what should be in that pack, before asking it to rapidly check every pack for those parts.
Another example would be a complex electronic assembly, like a smartphone. Using the same principle an AI would know what the correct part looks like and would quickly learn to find where differences exist, like missing screws, or loose fixings within the case. This could be also be used with returned goods to ensure that package are complete. This is more than just comparing two images and identifying differences. An AI would be able to detect if a part was present and in a different place and decide if that is, or is not a problem.
Beyond these narrow applications, we can teach our AI systems to do more than one task, broadening their intelligence. These broader systems are like our TruView Parts Counter AI, that is initially taught to count components on a reel from an x-ray image, but can then learn what a real component looks like and what a counterfeit component looks like. This broader level of AI or expertise starts to make the system behave more like a human, doing one simple task, in this case counting, while looking out for anything untoward, in this case fake parts. As we add more skills this broader AI system, it operates with greater speed and consistency becoming more like a super human good inwards inspector.
The future of AI in manufacturing
So, no dystopia, with AI powered robots running factories staffed by downtrodden humans answering only to a number. Instead super-augmented operators who work with AI and x-ray inspection systems, teaching and training them to undertake narrow tasks with incredible speed, accuracy and reliability.
The future of AI in manufacturing is finding the processes where it actually works and the machines that, with AI, can gain superpowers. We all want operators to have superpowers and be augmented, both physically and mentally. AI is the tool that will make that possible.
In this special report, we give a glimpse of the status and outlook of the chips, the software, and the uses of this emerging technology.
To explore the status and outlook of AI in greater depth, check out all the stories listed below that are part of an Apencore special report on AI.
It’s Still Early Days for AI
AI is still in its infancy with some of the most interesting accelerators yet to be disclosed, software still evolving, and benchmarks yet to get fleshed out and exercised.
AI Silicon Sprouts in the Dark
Deep learning has spawned work on a wide variety of novel chips, but the most interesting architectures have yet to be designed, let alone benchmarked.
AI Code Wags Hardware–Vigorously
Deep-learning models, frameworks, and techniques like reinforcement learning are moving faster than you can carve out paths in silicon.
Why I Joined MLPerf
A microprocessor analyst tells how he got involved in an effort to benchmark deep-learning systems, what he learned so far, and what he wants engineers to know about the work ahead.
Kamen Aims to Deliver AI to FedEx
The engineer behind the Segway and First Robotics Competition discusses his team’s current work for FedEx on a delivery robot using neural networks.
AI Trolls for Data Center Woes
Neural networks require time, expertise, and plenty of computer power, said a senior engineer reporting on a project at Hewlett-Packard Enterprise.
China Sees U.S. Ahead in AI
Your competitive position in the global rush to deep learning varies depending on where you sit, according to a story of recent articles in EET-China.