AI in the enterprise: AI may as well stand for automatic idiot – but that doesn't mean all machine learning is bad
- Reference: 1599498008
- News link: https://www.theregister.co.uk/2020/09/07/ai_debate_against_motion_mon/
- Source link:
During the week you can cast your vote using the embedded poll, choosing whether you're in favor or against the motion. The final score will be announced on Friday, revealing whether the for or against argument was most popular. It's up to our writers to convince you to vote for their side.
For our first debate, the motion is: [1]Artificial intelligence in the enterprise is just yesterday's dumb algorithms rebranded as AI .
This morning, Thomas Claburn [2]argued for the motion.
And now, arguing AGAINST the motion, is KATYANNA QUACH...
After years of hype, it’s easy to become cynical about AI. Elon Musk’s declarations that fully autonomous cars - ones that can operate with no human intervention - are just round the corner are laughable, especially when numerous Tesla drivers have died in Autopilot car crashes.
If semi-autonomous cars aren’t completely safe, what makes you think that vehicles that can drive by themselves are going to be any better? AI algorithms? Surely not. Experiments have shown that computer vision systems in self-driving cars can be tricked into [3]swerving across lanes by adding stickers onto the road. Also, let’s not forget that a [4]piece of black electrical tape tacked onto a 35mph (56kph) road sign fooled Tesla vehicles into speeding up to 85mph (136kph).
The problem with these so-called adversarial examples don’t just affect Tesla; they can impact all types of deep learning models. These vulnerabilities exist because machine learning algorithms aren’t very smart and can only perform specific tasks that they have been trained to do. Feed them an input that slightly deviates from the ones encountered during the training process and it’ll make mistakes.
There are all sorts of issues with modern machine learning that it’s no wonder you’re highly suspicious of the rising number of companies claiming to use AI in their products or services. And so you should be, but does that mean there are no companies out there that are really using machine learning at all? Of course not.
The smart assistant wars
Let’s start with one of the most well-known examples from Big Tech: Amazon, Apple, Alphabet, Facebook, and Microsoft. Each biz has invested in a voice-activated AI assistant in smart speakers in one form or another. Amazon has Alexa, Apple has Siri, Alphabet has Google Home, Facebook has Portal, and, well, Microsoft just decided to [5]stop rolling out Cortana for the Harman Kardon Invoke device.
Sure, there’s a good argument to be made that these gizmos aren’t all that useful and are terrible with [6]less common accents . Yes, abilities like reading from Wikipedia or setting a timer for the oven are hard-coded instructions rather than fancy AI algorithms. But there is still a touch of machine learning in the works.
Speech recognition and speech-to-text models are needed to kickstart these smart speakers into gear after all. A simple wake word like “Hey, Siri!” or “OK, Google” uttered by a user triggers neural networks to process the audio sample, and it gets ready to transcribe audio instructions. Natural language understanding algorithms help the assistants identify what function to perform based on the user’s speech. Big Tech has profited handsomely from these AI-powered speakers and have sold hundreds of millions of units.
Deep learning isn’t just limited to massive corporations with big bucks and large teams of engineers either. Smaller well-established companies and brand spanking new startups have successfully commercialised other areas of new-fangled machine learning.
Reinforcement learning is notoriously difficult to get working in the real world. Although the algorithms may have produced some of the flashiest results – think DeepMind’s [7]AlphaGo bot that bested the world’s leading Go player – they’re difficult to commercialise. But that’s exactly what [8]Covariant AI , a robotics upstart, has managed to do.
Earlier this year it announced that one of its robot arms was now helping sort through equipment like light sockets at a warehouse owned by Obeta, a German company that supplies electrical parts. “Even though we are just getting started, the systems we have deployed in Europe and North America are already learning from one another and improving every day," Pieter Abbeel, founder and chief scientist at Covariant, [9]previously said in a statement.
Sometimes market breakthroughs don’t come from flashy startups
Sometimes market breakthroughs don’t come from flashy startups. The first FDA-approved AI medical device that can diagnose eye diseases caused by diabetes doesn’t come from the likes of Google. The Chocolate Factory may have teams working on computer vision models to spot signs of diabetic retinopathy, a leading cause of blindness, in retinal scans. But it was beaten by IDx LLC, a lesser-known company based in Iowa, who now sell a product known as IDx-DR that automatically detects damage in the back of the eyeball from medical images.
AI has now progressed enough that cloud companies even offer off-the-shelf models to help companies perform simple tasks like translating languages or recognizing objects and faces. The difficult part of working out how to integrate these pre-built systems with your own data pipeline so that transferring data to and from the cloud to a particular product or service is seamless.
Although this might count as cheating since companies are offshoring the heavy lifting to cloud companies, you can’t deny that they’re still some form of machine learning that’s being offered even if it is being outsourced to Google Cloud, AWS, or Microsoft Azure. In these cases, it’s often that machine learning is just a small part of the solution and that most of the product or service being sold is still based on good ol’ fashioned engineering.
There is no magic trick when it comes to AI; it’s merely another software tool – albeit a very powerful one – that can be used by developers. Deploying machine learning models still requires traditional hardcoded techniques. Modern AI is used in commercial settings today, it’s just that these algorithms run quietly in the background cranking away at mundane tasks that are not normally nearly as exciting as what’s being hyped. And as they say, if it works it’s not AI. ®
Cast your vote below. You can [10]track the debate's progress here .
JavaScript Disabled Please Enable JavaScript to use this feature.
Get our [11]Tech Resources
[1] https://www.theregister.com/Debates/2020/09/07/ai_debate/
[2] https://www.theregister.com/2020/09/07/ai_debate_for_motion
[3] https://www.theregister.com/2019/04/02/tencent_tesla_hacking/
[4] https://www.theregister.com/2020/02/20/tesla_ai_tricked_85_mph/
[5] https://www.theregister.com/2020/08/08/ai_in_brief/
[6] https://www.theregister.com/2020/07/30/ai_assistants_uswitch_accents/
[7] https://www.theregister.com/2016/01/27/google_go_beats_human_master/
[8] https://www.theregister.com/2020/02/03/ai_roundup_310120/
[9] https://www.prnewswire.com/news-releases/covariant-launches-from-stealth-to-bring-universal-ai-to-robots-300995185.html
[10] https://www.theregister.com/Debates/2020/09/07/ai_debate/
[11] https://whitepapers.theregister.com/
1. Machine Learning is not AI
2. The training data is the code, and the "AI" algorithm essentially compiles this into assembly language that the computer understands.
If there are problems with the training data, then the algorithm won't work.
Problems with the training data might not always be the most obvious ones. For example, there was an attempt to train a computer to recognise a medical condition from viewing images. The doctors who supplied the images helpfully circled the areas which indicated the medical condition was present. A human trying to learn from these images would understand that this was a circle added by the doctor, and concentrate their efforts in looking at the area highlighted by that circle. The computer just decided to look for images with circles drawn on them, because it was too dumb to understand that there were circles that were not part of the original image.
Computers are no more intelligent now than they were in the 1970s when Unix came out. Most of them still run Unix, or something based on the Linux kernel which is basically the same as Unix. Windows came later and is different, but not in any way that is relevant to intelligence.
Matt Parker has a series of videos on YouTube about using machine learning to train a computer to learn noughts & crosses (en:US tic tac toe). As noughts and crosses is a very simple game, the ML model is small enough to understand what is going on. It can, with suitable training, learn how to play one perfect game, but it doesn't understand the different strategies you can use (eg. corner / centre / edge), and how to deploy them to your advantage.
"1. Machine Learning is not AI"
And there you have it, stop calling machine learning, deep learning and neural networks intelligent , they are not, the use of the term Artificial Intelligence for any of the current uses at the current state of the technology is fraudulent.
Without doubt there are machine learning algorithms that are extremely useful, particularly in industry and medicine.
Are they clever? Yes.
Are they intelligent? No!
Deploying machine learning models still requires traditional hardcoded techniques
What are traditional hardcoded techniques?