AI-Washing: How to Spot It and Why It Matters

As artificial intelligence (AI) becomes a dominant buzzword across industries, more companies are rushing to label themselves as AI-powered. However, many of these claims don’t hold up to scrutiny. This growing trend, known as "AI-washing," involves exaggerating or misrepresenting the role AI plays in a company’s products and services. While genuine AI solutions leverage machine learning (ML) and deep learning to adapt and improve over time, many so-called "AI-powered" tools are simply using older technologies, such as rules-based algorithms or model-based analytics, that lack the core components of true AI.

The Blurring Lines: AI vs. Other Technologies

Understanding the difference between AI and traditional methods is key to identifying AI-washing. The following technologies are often labeled as AI, even though they do not meet the criteria for what true AI involves:

• Model-Based Analytics: This refers to statistical models used to interpret data and generate insights based on predefined parameters. While effective for some tasks, these models do not autonomously learn or adapt, as machine learning models would. For example, companies that provide predictive analytics for stock prices or customer churn often use static models that can only be as accurate as the data and assumptions they were originally built on.
• Rules-Based Algorithms: These systems function using a series of explicit, human-defined rules. These rules drive decision-making based on if-then logic but do not allow the system to learn from new data. Expert systems in fields like law or medicine that follow pre-established guidelines are examples of rules-based systems that may be passed off as AI. While they mimic intelligent behavior, they lack the adaptability that is the hallmark of true AI systems.

Example: GNU Chess vs. Deep Blue

To highlight the difference between rules-based systems and true AI, consider the example of GNU Chess versus Deep Blue. GNU Chess, an open-source chess engine, operates on predefined heuristics and rules. Its decisions are based on algorithms coded by humans, and it does not learn from experience. Contrast this with IBM's Deep Blue, the AI that famously defeated chess grandmaster Garry Kasparov in 1997. Deep Blue incorporated machine learning, which allowed it to analyze positions, predict outcomes, and improve its strategy over time.

Real-World Examples of AI-Washing

1. Financial Services

The financial sector has been particularly guilty of AI-washing. Many fintech companies promote their offerings as AI-driven, whether it be for investment recommendations, fraud detection, or customer service. However, many of these companies use simple regression models or rules-based decision trees, which are not AI but rather traditional methods of data analysis.

• Robo-advisors, like those from companies such as Betterment or Wealthfront, often claim to use AI to optimize investment portfolios. However, most rely on fixed algorithms that rebalance portfolios based on pre-determined criteria such as risk tolerance, rather than any real-time learning from market conditions.
• Fraud detection systems: While some banks like HSBC and JPMorgan Chase have advanced their fraud detection models with machine learning, many smaller institutions market traditional anomaly detection algorithms—based on historical data and human-set thresholds—as AI-driven fraud protection. These systems flag transactions that deviate from predefined patterns but lack the ability to learn and adapt to new types of fraud schemes.

2. Healthcare and Biotech

The healthcare industry, in particular, is seeing a flood of companies claiming to use AI to enhance diagnostics, treatment planning, and drug discovery. While AI holds incredible potential in these fields, many products being marketed as AI-powered are little more than enhanced data analytics.

• Theranos: Although a now-infamous example of fraudulent claims rather than AI-washing specifically, Theranos provides a cautionary tale. The company claimed to revolutionize blood testing using breakthrough AI-powered diagnostics. However, the technology did not exist. The case illustrates the dangers of unchecked hype in healthcare innovation.
• AI in radiology: Numerous startups claim their tools use AI to detect diseases like cancer from X-rays and MRIs. For example, companies like Aidoc and Zebra Medical Vision promote AI-based image analysis, but not all products in the space use machine learning or deep neural networks. In many cases, they are relying on rules-based image recognition algorithms, where specific patterns are pre-coded by human experts rather than being learned by the model from vast amounts of data.

3. Retail and E-commerce

E-commerce platforms often market their recommendation engines or customer service chatbots as AI-powered. While AI can significantly enhance personalization and improve user experience, some companies use simple, rules-based filtering and keyword detection to replicate what AI could otherwise accomplish.

• Chatbots: Many companies claim their chatbots are AI-powered, but in reality, they are merely following pre-set response scripts. For example, customer support chatbots often use natural language processing (NLP) to match user queries to relevant responses from a database, but lack the dynamic learning and conversation capabilities seen in more advanced AI systems like GPT-4 or Google's LaMDA.
• Recommendation engines: Retail giants like Amazon have successfully implemented machine learning-based recommendation engines, where the system learns user preferences over time and adapts its suggestions. However, many smaller e-commerce companies simply use basic collaborative filtering (if you bought X, you’ll likely want Y) that operates off simple rules rather than learning algorithms.

4. Autonomous Vehicles

The autonomous vehicle sector is another hotbed for AI-washing. Companies that are developing autonomous driving systems often exaggerate the role of AI, when in fact many are still reliant on rules-based automation rather than true autonomous AI.

• Tesla has faced criticism for marketing its Autopilot feature as “Full Self-Driving” (FSD), leading some to believe that the cars are AI-driven and fully autonomous. However, Tesla’s Autopilot, while impressive, is still heavily reliant on pre-programmed driving rules and does not yet qualify as Level 5 autonomy, where the vehicle would need no human intervention. In contrast, companies like Waymo are using more advanced AI techniques, including machine learning to recognize and predict the behavior of other drivers and pedestrians.

Spotting AI-Washing

With AI-washing so prevalent, it’s crucial to develop a discerning eye for which technologies are truly AI-powered and which are being exaggerated. Here are a few key questions to ask when evaluating a company’s AI claims:

1. Does the system learn and improve autonomously? Real AI involves machine learning or deep learning, where systems learn from data and adapt over time. Static models or systems that require regular human intervention are not AI.
2. Is there clear transparency in the algorithm’s function? True AI models are complex, often utilizing neural networks and adaptive learning techniques. Ask for details on how the system’s model works—if it’s rules-based or manually programmed, it likely isn’t AI.
3. Are AI capabilities solving non-routine, evolving problems? AI excels in environments that are uncertain or where solutions require real-time adaptation, like autonomous driving or real-time customer service. Tasks that follow predictable, repetitive patterns are often better handled by traditional algorithms.

Why AI-Washing Hurts Innovation

AI-washing not only confuses customers but also stifles genuine innovation. By overstating the role of AI in products, companies create unrealistic expectations, leading to disillusionment with the technology when it fails to deliver. This undermines the real advances being made in the field by companies that are using AI to tackle complex, evolving problems.

Spotting AI-washing is more than just cutting through the hype—it’s about promoting transparency and supporting companies that are truly pushing the boundaries of AI. As AI continues to evolve, ensuring that we maintain clear distinctions between AI and other data-driven technologies will be essential for fostering trust and innovation in the market.