SafetyDetectives recently spoke with Roei Ganzarski, CEO of Alitheon, to delve into the cutting-edge world of optical AI and machine vision technologies. Ganzarski provides an enlightening perspective on the unique challenges and innovations his company has navigated in the realm of product serialization and authentication and supply chain security. Drawing from his rich background in leadership roles across various technology sectors, he highlights how past experiences have finely honed his approach at Alitheon. This interview explores the misconceptions surrounding AI and blockchain within the industry and sheds light on the practical applications and limitations of these technologies in combating counterfeit and gray market products. Ganzarski’s insights offer a glimpse into how Alitheon uses machine vision in enhancing the integrity and safety of products across global markets.
Can you share a bit about your journey before Alitheon and how your past experiences have shaped your approach as CEO?
My name is Roei Ganzarski, and I currently serve as the CEO of Alitheon. Before joining Alitheon, I was the CEO of magniX, an electric propulsion company for aircraft, and simultaneously the Executive Chairman of Eviation Aircraft Ltd. Prior to that, I led a software company called BoldIQ, and I also spent a significant amount of time at Boeing here in the Seattle area.
My journey to Alitheon has been a series of stepping stones, each adding to my understanding of what to do—and what not to do—in leadership roles. These experiences have taught me about effective approaches and the pitfalls of certain rules, regulations, policies, and processes. It’s not just one specific event but rather the cumulative learning throughout my career that has shaped my approach as a CEO.
Can you tell us about Alitheon and some of your flagship features?
Alitheon is an optical AI company, specifically, a machine vision company that has developed technology to “fingerprint” objects. Fingerprinting for people is popular with governments and other entities because it doesn’t rely on proxies that can be manipulated or faked—like driver’s licenses, badges, tags, or passports, which can be altered, forgotten, or impersonated.
Human fingerprints have two crucial attributes:
- Uniqueness: Even twins and triplets have different fingerprints.
- Inherent and Persistent Nature: They are a part of your body, meaning they can’t be forgotten or transferred to someone else, making impersonation extremely difficult.
That’s why authorities check your fingerprints at airports or government buildings—they are your unique identifiers. We’ve developed a system that applies the same principle to objects. Every item manufactured, regardless of whether it is made of plastic, wood, or metal, and regardless of the manufacturing process like 3D printing or CNC milling, has unique, inherent flaws due to production tolerances.
Design and mechanical engineers might specify, for example, that a part should be three inches wide with a tolerance of a thousandth of an inch. Though machines can’t produce exactly to these specifications every time, they achieve a quality within that tolerance that appears consistent to the end user.
What we’ve done at Alitheon is look within that tolerance band to identify unique features of manufacturing—flaws or characteristics that aren’t added intentionally but arise naturally from the manufacturing process. We can detect these with standard cameras, and they act like a fingerprint because of their uniqueness—one in 3.5 trillion chance of repetition in another product.
We’ve found a way to capture this as a digital “FeaturePrint” of a product by simply taking a picture with a standard camera, which could be an industrial or even a smartphone camera, without special lighting or marking the item in any way. There’s no need for system training; our algorithms automatically identify the item whether it’s a pen, a watch, or a gold bar.
When someone takes a picture of that item again, our system performs a match—similar to what you see in shows like “CSI”—to verify its identity. For example, it will confirm whether it’s bottle number 176-3, (not 176-2 or 176-4), precisely the one you bought last week, and not another one that is seemingly identical.
This capability allows our customers to trace their products—knowing where they were last seen, who last handled them, and where they were installed, as examples. It helps prevent counterfeiting, as a counterfeit would not be registered in the system. It also helps eliminate gray market items—authentic items sold illegally—such as expired medicine with altered dates, or items that failed quality control but are sold as if they passed, stolen items, or items manufactured without the company’s permission. By simply identifying items accurately and irrefutably, you can secure and ensure your supply chain, and protect against theft and fraud.
What have been some of the most significant challenges in developing and deploying optical AI technologies?
Developing our technology presented several significant challenges before we were ready to go to market:
- Utilizing Standard Cameras: We aimed to avoid requiring our customers to purchase specialized equipment, like expensive spectral imaging cameras or x-ray cameras, which would be impractical for widespread use. Our goal was to make our technology compatible with standard, off-the-shelf cameras, including industrial and cell phone cameras. The challenge was to identify tiny and invisible features with such basic equipment.
- Product Identification Without Prior Knowledge: It was crucial for our technology to identify unique features of any product without prior knowledge of what that product might be. For example, a water bottle and an automotive brake pad have vastly different characteristics. Our system needed to reliably detect anomalies or specific features pertinent to each unique item, without the need for prior ‘training’ of the system.
- Eliminating False Positives: Achieving zero false positives in production, was essential, especially in critical fields like pharmaceuticals, aviation, and automotive. The stakes are incredibly high; a false positive—misidentifying fake medicine as real, for example—could be life-threatening. Thus, ensuring the reliability of our identifications was a top priority.
We spent approximately five years developing our deep technology to overcome these challenges successfully. Now, we bring to market a technology solution that meets these stringent requirements.
How do you see the role of AI evolving in product authentication and supply chain security sectors?
To begin with, it’s important to clarify what we mean by AI. In our context, AI refers to machine vision, but it’s not just about artificial intelligence—it’s about augmented intelligence. Our goal is to enhance human capability to make decisions that are otherwise difficult, if not impossible, due to limitations in human perception and abilities. For example, distinguishing whether a product is genuine or counterfeit, or determining ownership when two items appear identical to the naked eye, are tasks that humans alone, without the right tools, cannot reliably, consistently, and objectively perform. Our optical AI system empowers users to make these distinctions with certainty.
We describe this as augmented intelligence, utilizing machine vision to significantly reduce, if not completely eliminate, counterfeits, gray market goods, and mistakes, while making item traceability a simple given. The increase in fraudulent activities, driven by human greed, poses significant risks. For instance, counterfeiters might replicate luxury goods like watches or bags, which is illegal and harmful to brands, but not typically life-threatening. However, the stakes are much higher when it comes to counterfeit brake pads, medications, or airplane parts—where the falsification directly endangers lives.
Our goal is to help our customers avoid that, or at least provide companies and consumers with the tools to ensure it doesn’t happen. If the tools are there to guarantee that counterfeiting can’t occur—meaning we can catch any fake part that goes on a plane—then that should discourage people from attempting to counterfeit critical components, because they know they’ll get caught. So it could deter fraudulent activity. That’s the ultimate aim, and it highlights the role of augmented intelligence in supply chain manufacturing.
What are some of the biggest misconceptions businesses and consumers have about AI and machine vision in the context of product authentication and safety?
There are a few misconceptions. One major misconception is about blockchain. We are not blockchain; we don’t use blockchain. Some of our customers do for other reasons, but blockchain will not protect your supply chain or your products. Blockchain is a phenomenal tool for protecting the integrity of the digital data you put into it, but that’s it. If the data is fake or wrong, the blockchain doesn’t make it real or right; it’s still fake. It might confirm that the data associated with the product hasn’t been manipulated, but it doesn’t mean this is actually the real and specific product. This leads to the misconception that if you use blockchain, you’re protected. No, you’re not protected; your data is protected. Your physical products aren’t.
Another misconception is that there is some sort of magical AI that, without having seen the original product and taking its fingerprint, will find out what it is. This would be the equivalent of people thinking, “I found a fingerprint at a crime scene and AI will find out who it belongs to immediately, without ever having the person fingerprinted beforehand.” No, you have to have the baseline fingerprints in the system. If a criminal’s fingerprint is found at a crime scene and that person has never been fingerprinted before, you will never identify them. People get excited about the notion of AI and take it so far that it becomes dreamland.
For instance, we can identify things that we’ve registered, and then people say, “Oh, that’s great. Can I show you the other side of the product and you can identify it?” The answer is no, just like with a fingerprint. If I registered the right finger, you can’t show me the left finger and expect the system to know who I am; it has to be the same finger.
Similarly, you can’t say, “AI will tell me if the sneakers are real or if they’re gray market.” Now, machine learning can maybe identify an authentic sneaker from a fake one, if that specific sneaker (both real and fake) has been trained into the system. It’ll learn the slight differences and tell you, “I think that’s real” or “I think that’s fake.” That’s doable technically, but it doesn’t solve gray market issues, human error, or traceability. If those sneakers are real but stolen, or were made in a ‘midnight factory run’, AI or machine learning won’t solve that. It will tell you, “These are real sneakers,” but it won’t tell you, “These are the stolen ones from your factory,” or “the wrong ones to use” because they’re not serialized.
Another misconception is that AI can serialize; AI can work at the class level. For example, going back to the fingerprint, if you went to a crime scene and found a fingerprint, the AI system would say that was a human being. Now, that’s nice to know—it was a human and not a bear or wolf—but you really want to know which human being. That’s the whole essence of the fingerprint, and that’s what we do for products.