Artificial intelligence has a vocabulary all its own. Just within the field of machine learning, you’ve got a bevy of terms and concepts to sort out: supervised versus unsupervised ML, deep learning and neural networks, and black box versus explainable AI.
It’s a lot. If you need to brush up on the lingo fast, try our AI cheat sheet. For a deeper dive, try the executive’s guide to real-world AI.
Because of this – and because of the outsized enthusiasm and hype that engulfs AI – there is some fundamental confusion around AI and related technologies. What separates AI from other forms of automation? What qualifies a tool or service as a full-fledged AI product – rather than something that simply leverages AI or other automation? And while we’re at it: Why does it matter?
“This question is an interesting one and comes up often because we are starting to see AI all around us – whether we notice it or not!” says Partho Nath, head of applied AI at Netomi.
Indeed, that’s one reason why certain distinctions do matter: AI is everywhere, or it will be soon. AI/ML ranks as the top emerging technology workload in Red Hat’s 2022 Global Tech Outlook, with 53% of IT leaders reporting plans to use it during the next 12 months, a three-point increase from the previous year.
[ Related read: Automation: 5 expert tips to advance your journey. ]
Meanwhile, 56% of respondents in McKinsey’s 2021 State of AI report said they’ve already adopted AI in at least one business function, up from 50% in 2020.
As adoption increases, so does the volume of vendor pitches – not just for AI technologies themselves but other tools that use – or claim to use – AI in their solution. (Such claims are worthy of closer inspection. One study – conducted back in the halcyon days of 2019 – of more than 2,800 European startups that claimed to be “AI companies” found that roughly 40% of them weren’t actually using AI in a substantive way.)
While the average end-user might not even notice – nor need to notice – some of the finer distinctions, IT leaders and their teams have a different relationship with AI. Developing criteria for evaluating different technologies is, as always, part of the job. Here are some ways to boost your own thinking and discussion on the subject.
What is an AI product?
As a high-level term, an “AI product” clearly implies AI as the central technology in an application or service – rather than something that is merely “AI-like” or an aspirational claim about what something might be able to do at some undefined point in the future.
There are two good rules of thumb that can guide your own definitions and analysis:
- Can it perform work like a human would?
- Can you develop and customize it to support your specific requirements (turnkey, one-size-fits-most solution?)
“As a CIO, I look at an 'AI product' as one that can introduce a human element into a process,” says Carter Busse, CIO of Workato. “Meaning, if the product can perform a task that normally a human would – like troubleshooting a customer or employee issue – or it can initiate an automated process from a simple question, that’s an AI product.”
The distinction Busse makes here explains why, for example, robotic process automation (RPA) is not an AI product. RPA is good at rules-based, if-then tasks but it can’t think for itself or adapt to changing conditions. That doesn’t mean RPA isn’t valuable – it just means it’s not AI.
It also illuminates another reality: AI is hard work. (This is another reason why sales pitches that make AI sound as easy as an ocean breeze should probably be viewed with suspicion.) Getting to the point where the AI and underlying algorithms can handle a process as a human would requires both an up-front investment of time and a long-term commitment to continuous optimization.
“The AI process is usually learned over time using ML algorithms that are constantly being tweaked to make those human-like interactions possible,” Busse says.
The ability to customize – to make those tweaks Busse describes, but also to generally develop for your goals, rather than fitting your goals to preset, out-of-the-box capabilities – is the second key piece.
“The key differentiation between something that is truly an AI product versus something that simply leverages AI/ML comes down to the degree of customization supported in the platform, right down to the engine itself,” says Nath from Netomi.
Tools that fall into the latter category often use off-the-shelf libraries to embed some AI capabilities in a broader solution. Nath notes that this is perfectly fine in scenarios with AI/ML just “a cog in the wheel” and the actual focus is something else.
An AI product, on the other hand, features “AI as the centerpiece and the main decision-maker,” Nath says. “These products often have a large degree of customization options so that the AI can be fine-tuned for the specific use cases where it is being deployed.”
What’s the difference between an AI/ML product and an AI/ML model?
AI in the enterprise today commonly means that an organization is running machine learning models in production.
A model – or even dozens or hundreds of them, in the case of more advanced teams – is not a product, per se. But what you do with the outputs of that model can become an AI product.
“AI and ML can analyze existing data or generate new data,” explains Michael Roytman, chief data scientist at Kenna Security, now part of Cisco. “What’s done with that analysis and how it fits into the ecosystems customers live in is a question solved by products, not models.”
Roytman describes AI products are fully integrated extensions of an AI or ML model and uses a universal app – weather – to illustrate the difference between an app that relies on an ML model and a fully realized AI product.
“A service that looks at existing weather data and makes a prediction about the average rainfall is certainly an ML model – likely a regression or a neural network,” Roytman says. “An AI product would include the dashboards, workflows, likely integrations, and necessary user controls.”
When is AI actually "just" automation?
A related concern here is that AI sometimes gets used interchangeably with automation. AI is a form of automation and an automation enabler, but there are many other types of IT automation that aren’t AI – so the terms shouldn’t be used as synonyms.
“From a technological point of view, it is easy to differentiate an AI from an automation-based product,” Nath says. “Automation-based software is only able to respond to outcomes it has been programmed for.”
[ On the IT job market? Read IT jobs: 7 hot automation skills in 2022. ]
That describes RPA, for example, as well as other forms of automation like infrastructure-as-code and container orchestration. These are incredibly valuable technologies, but they’re not AI.
Busse, the Workato CIO, draws the line between AI and other forms of automation with data intelligence.
“Data is useless if a company doesn’t utilize it the right way, so it’s integral that an AI product uses data intelligence to inform decisions that real-life people would otherwise be responsible for,” Busse says.
This is IT irony: A defining trait of an AI product is the ability to perform work that would otherwise need to be done by a human, yet this ability depends on continuous human oversight for optimal performance.
The overarching goal of other types of automation should be to put yourself out of a job – not literally, per se, but in the more figurative sense usually meant by sysadmins, SREs, and other automation-focused IT pros.
“Automation, on the other hand, follows pre-programmed rules and runs with little to no human interaction,” Busse says. “Once the specific patterns – usually for repetitive tasks – have been identified and arranged in that piece of software, the need for human contact after is minimal.”
[ How does AI connect to hybrid cloud strategy? Get the free eBook, Hybrid Cloud Strategy for Dummies. ]
