The Real Cost of AI: Debunking the Myth of Free Artificial Intelligence
Welcome to the first newsletter of 2023: a year I think could be one of the most groundbreaking for AI, and it's not even about tech (because while it might surprise you, tools similar to? ChatGPT have existed for some time).?
It's about the sheer interest in AI and the growing awareness and attention being given to this powerful technology. But what will I write about this month? Well, here’s what we’ll cover today in today’s edition:
When is custom AI not the right move?
The world has gone AI-crazy. Companies want to implement AI because they know it can bring immense benefits. As the CEO of a company that has been doing this for eight years, I really see the explosion of interest.?
Still, familiarity with AI varies considerably. We often hear from companies with a vision, product demos, even pre-prepared models for implementing an AI-based project. Such companies know the market well and typically have strong technical skills on their team.
On the other hand, some companies are trying to reinvent the wheel ?? (by which I mean they’re looking to build AI to solve some internal problem when an existing tool could do it for them).
If profit was my only driver, I could build something new and say nothing about these other tools. However, at DLabs.AI, we prefer to help companies tailor solutions to their needs. That's why, on several occasions, we’ve suggested clients try a ready-made solution.
Some of you probably want to know the kind of tools I’m talking about? Well, it depends on your business objective, but here are some examples:
And always be mindful: some tasks are just as easily handled in Google Sheets! But whatever you’re after, the site There's AI for that is a great place to start :)
Why is AI so expensive?!?
Another time it doesn’t make sense to invest in AI is when you’re looking to do something for the lowest price possible.?
The last few months have involved some intense client conversations, each proving invaluable to the team and me. One thing I’ve seen is a lack of awareness among business owners of the complex, time-consuming and costly nature of creating AI from scratch.
Perhaps this is partly due to the advent of 'free' tools like ChatGPT and Midjourney. But the truth is: if you want to integrate GPT into your product, it isn't free (you have to pay OpenAI to access the API), and developers have to integrate. That’s why the salaries in the tech sector are growing.
And I'm not surprised - when you have to invest years of your life in learning math, computer science, or other science subjects, you want this to pay off.
Anyhow, back to the costs of building from scratch. Let’s look at some Gartner data, which tells a story in itself:
That said — there are ways to reduce the risks. Failure is much less likely if you work with an experienced AI developer.?
If you find a professional company, you’ll also get sound advice on how to ensure the product meets both technical and KPI expectations. Moreover, you’ll end up with the latest tools, and your teams will feel well prepared to integrate AI into their workflows.
But that’s not all. The best partners know how to use AI to unlock new business opportunities (and even identify the one with the highest ROI), helping you find interesting ways to further monetize your existing products or services.
Finally, when you work with people who have tackled similar projects before (and so have seen the potential problems and pitfalls), you minimize the risk of failure, which, in turn, avoids you burning your budget.
This is perhaps the biggest hidden benefit of working with an experienced AI team, and it’s hard to put a monetary value on this item. Still, how much should you budget overall??
Well, that depends on the following:
That’s just the tip of the iceberg. But let's answer the BIG question, "Just why is it sooo expensive?!?"?
The primary cost is time. Depending on the project scope, several specialists (including data scientists, machine learning engineers, project managers, software developers, and others) will have to work on it.
Sure, you could try to hire these people yourself. But there’s a significant cost associated with recruiting, onboarding, and training these kind of personnel (which could turn into an ongoing headache if an employee were to leave your project halfway through).
领英推荐
When you work with a specialist provider, they are adept at handling such challenges. And you only ever pay for the hours the team spends on your project, not covering additional expenses of hiring full-time employees.
So even if you feel the hourly rate of a development partner seems a little high, just remember how much you’re saving by outsourcing all the other elements, too.
Beyond personnel, there are several other costs to consider, including:
And… depending on the project, there may be other expenses, including data collection, annotation, legal fees; the list is pretty extensive.?
That’s why we try to be as realistic as possible when pricing new projects, as we don’t like raising the quote down the line. And being clear upfront means our clients know exactly what to expect.
Interested to learn more? Check our article on ‘How Do You Estimate The Time And Cost Of A Machine Learning Project?’
AI needs just 3 seconds to mimic your voice
ChatGPT may be the talk of the internet. But my attention has been on another exciting tool called VALL-E, recently announced by Microsoft researchers.?
VALL-E can accurately simulate a person's voice using just a 3-second audio sample. And once it’s learned a voice, it can synthesize the sound of that person saying anything — and do it in such a way as to preserve the speaker’s emotional tone.?
The team behind VALL-E speculates that, in combination with other artificial intelligence models, it could be used to create high-quality applications for text-to-speech conversion, speech editing, and audio content creation.
The team trained VALL-E on an audio library called LibriLight, which contains 60,000 hours of English speech from more than 7,000 speakers. And the voice in the three-second sample must closely match the voice in the training data to get a good result.
Microsoft gives dozens of examples on the VALL-E website showing the AI in action. The results are impressive: in some cases, the two samples are almost indistinguishable. That said, some results do sound computer-generated, even if many could be mistaken for human speech.
You may now be worrying that criminals could use it to impersonate specific people. Fortunately, the developers are aware of this risk and are working on a detection model to distinguish whether VALL-E has synthesized a particular audio clip.
Source: ARS Technica
Using AI in medical imaging is saving lives
Last but not least — let’s dive into AI in healthcare.?
This month, I’ll focus on AI in medical imaging. As you might know, technology has been helping doctors analyze medical images for years. But did you realize how much of an impact it’s now having?
See just a handful of the astounding results below:
Unfortunately, processing medical images is still a significant challenge, with the problem being the sizable input formats.
Tissue samples are often digitized in ultra-high resolution. Meaning the file size of these images can be several gigabytes, which makes them impossible to load in a generic image viewer (due to a lack of memory to accommodate a deserialized image).?
Can we solve this problem? Well — our team just worked on an interesting saluting during a Mayo Clinic – STRIP AI competition, which was focused on using image classification to identify a stroke blood clot origin.?
The goal was to classify the blood clot origins in an ischemic stroke. And using whole-slide digital pathology images, participants had to build a model that differentiated between the two major acute ischemic stroke etiology subtypes: cardiac and large artery atherosclerosis.
This article by Tomasz Ma?kowiak (Machine Learning Engineer at DLabs.AI) covers the solution in detail, describing how to efficiently process large medical images using Apache Beam.
—----
And so we come to the end!
Thanks for reading my first newsletter of 2023; I’m so grateful for your continued support.
Now, see you in February ??