AI in the Real World: Navigating its Potential and Pitfalls

I recently completed a course on 'Leveraging AI for Business Success and Social Good' as part of my 2nd year curriculum at BITSoM - BITS School of Management . It was very enlightening and captivating as I learnt the profound impact of AI as a transformative force, reshaping industries and revolutionizing the way businesses operate.

One of the key takeaways from the course was - 'AI is not without imperfections'.

To explore this further, we dissected three prominent cases – Zillow, IBM Watson and the British Medical Journal (review of the Covid-19 diagnosis / prognosis) to unravel the complexities and lessons for aspiring enterprises.

Lessons from Zillow

Zillow's ambitious foray into the world of AI and real estate 'seemed like a dream come true. It worked on a simple concept - predict home values accurately, make competitive cash offers, and make modest renovations for a quick sale. This was ‘supposed’ to be a win-win for Zillow and homeowners, offering a seamless, efficient process that bypassed the traditional real estate market.

Zillow initially seemed to be doing everything right and signaled the following key takeaways for aspiring enterprises :

• Embrace Data as the foundation : Zillow emphasized on the importance of centrality of data. It was an example of how AI driven data analytics can redefine an entire industry. By collecting, organizing, and analyzing vast amounts of real estate data, Zillow provided valuable insights to sellers, homebuyers, and real estate professionals. Enterprises should recognize the immense potential of data and prioritize data infrastructure investments.
• Predictive analytics for business growth : Zillow also emphasized on the effectiveness of predictive analytics in improving customer experience and decision-making. It gives users incredibly useful information by predicting market trends and property values using AI and machine learning models. Enterprises should invest in predictive modelling to foresee market developments and proactively adapt further.

However, Zillow's journey with AI provides a cautionary tale.

In an article on Medium, Pedram Ataee, PhD mentioned “AI fails when history doesn't rhyme”.

While Zillow initially seemed to be doing everything right, the volatility that it experienced in the labour market or supply chain, made the prediction near impossible compared to the times before the pandemic, directing towards the following risks of AI based implementation:

• Overreliance on Data - One of the key pitfalls was Zillow's overreliance on its algorithms. The AI models, supported by vast amounts of data, were designed to predict housing prices with pinpoint accuracy. However, the risk here was that these algorithms did not account for sudden market shifts or unprecedented events, like the COVID-19 pandemic. This led to unpredictable buyer preferences. Enterprises should recognize the limitations of AI based models in handling unpredictable factors and have contingency plans in place to respond to such events.
• Operational and Scalability Challenges : Zillow encountered challenges in managing and renovating homes at scale, which led to increased costs and difficulties in turning a profit. The complexity of scaling AI-driven operations can be a risk if not managed effectively. Therefore, it is important for enterprises to have a well-thought-out scalability plan in place before implementing AI-driven operations at scale.
• Technology Limitations : The risks of AI in industries like real estate became all too apparent, revealing that while technology can significantly transform the industry, it cannot completely replace the human judgment nuances, and uncertainties involved in buying and selling homes. As industries are constantly evolving, it is important for enterprises to stay updated and reassess AI models and strategies as markets and consumer behaviours change.

Lessons from IBM Watson?

“IBM’s artificial intelligence was supposed to transform industries and generate riches for the company. Neither has panned out” – New York Times

A decade ago, IBM's Watson wowed the world by beating a human at "Jeopardy!" It seemed like AI was the future, ready to transform industries. Here's what it taught aspiring businesses:

• Strategic AI Integration : IBM Watson illustrated the importance of strategically integrating AI into a business's operations. Enterprises should consider AI as an essential component of their overall business strategy rather than as a stand-alone endeavor. Enterprises can make sure that AI investments generate measurable benefit and are in line with their long-term vision by coordinating AI projects with their primary objectives.
• Adaptation and Resilience : Watson's story emphasizes that the journey of AI in the mainstream economy is more about step-by-step evolution than a revolution. It demonstrates the importance of adaptability and resilience. Enterprises can learn that incremental progress, adaptability, and learning from challenges can lead to long-term success.

For many people, Watson became synonymous with AI itself.

However, Watson's journey was marked by missed opportunities and unmet expectations.

In an article in the New York Times, scientist David Ferrucci warned that Watson wasn't a one-size-fits-all solution ready for business, and it might even struggle with a basic second-grade reading test.

Let's explore what we can learn about AI risks from IBM Watson's story

• Gap Between Technology and Business Understanding: Watson's early success in a trivia game led to the misconception that it could tackle any commercial challenge. However, the technology was powerful but limited. This gap between technology expertise and business leadership can lead to misguided efforts and wasted resources.
• Complexity of Real-World Applications: One of the central risks in AI is underestimating the complexity of real-world applications. Watson's initial foray into healthcare, for instance, revealed the intricacies and messiness of medical data, which clashed with Watson's capabilities. Aspiring enterprises should understand the complexities and nuances of real-world applications when developing and implementing AI solutions.
• Costly Failed Initiatives: Watson's journey comprised of costly initiatives that promised significant returns but ended in failure. A project aimed at creating a diagnostic tool for cancer treatments, in collaboration with MD Anderson Cancer Centre, consumed $62 million and resulted in frustration and disappointment, emphasizing the high-risk nature of overambitious AI projects. Aspiring enterprises should establish realistic goals and identify tangible, achievable milestones for AI initiatives.

Lessons from the British Medical Journal

The British Medical Journal (BMJ) research paper study is primary focus on the clinical aspects of the COVID-19 diagnosis / prognosis and offers the following valuable lessons and insights regarding the application of artificial intelligence (AI) in healthcare and beyond:

• Speed and Scalability: The research points out that the COVID-19 pandemic generated an unprecedented amount of data, which was analysed quickly. AI and machine learning played a crucial role in developing and validating these prediction models in a timely manner. Aspiring enterprises can learn from this that AI offers the ability to analyse large datasets rapidly (scalability), even in high-pressure situations like a pandemic (speed).
• Data-Driven Decision-Making: The research emphasizes the importance of data for building and validating prediction models. It underlines the significance of comprehensive and high-quality data in developing robust AI models. Aspiring enterprises can learn that data-driven decision-making and data quality are fundamental to the success of AI projects. Collecting, managing, and using data effectively is key to building accurate and reliable models.

The research also highlighted the pitfalls of undertaking ambitious AI projects without due diligence. The insights from the BMJ research paper present a cautionary tale on the risks of AI:

• Data Quality and Bias: One of the prominent risks highlighted in the paper is the potential for data bias. Many models used data from a single country, and in some cases, data sources were unclear. This limited dataset diversity can lead to biased results, and similar issues can occur in AI models if they are trained on non-representative or biased datasets. Enterprises should ensure diverse and high-quality data is essential for reliable AI predictions.
• Model Overfitting: The paper mentions that small sample sizes were common among the studied models, increasing the risk of overfitting. Overfit models perform well on the training data but generalize poorly to new data. AI models can suffer from overfitting, especially if there is insufficient data. This can lead to inaccurate predictions in real-world scenarios. Aspiring enterprises should prioritize gathering high-quality and abundant data and recalibrate AI models.
• Real-world Variability: The paper discusses the variability of model performance across different settings and regions. This highlights the importance of considering the real-world variability and adaptability of predictive models, especially in the context of AI for healthcare, where local factors can affect model performance. The evolving nature of COVID-19 and healthcare more broadly underscores the risk of AI models becoming quickly outdated. AI systems need to adapt to changing circumstances and continuously update their knowledge and algorithms. Relying on static or outdated models can lead to misinformed decisions

The Way Forward

The learnings from case studies of Zillow, IBM Watson, and the British Medical Journal underscore the importance of embracing data, strategic integration, adaptability, and the complexities of real-world applications in AI projects.

These case studies underscore the dual nature of AI as a powerful tool with remarkable potential, yet fraught with risks and limitations. Enterprises must tread carefully, balancing the promise of AI with a deep understanding of its constraints, to harness its true transformative potential while mitigating potential pitfalls.

Thank you for reading. As we continue to navigate the ever-evolving landscape of artificial intelligence, join the conversation with these relevant hashtags: #AIInnovation #TechTransformations #BusinessAI #BITSoM #AISuccess #DigitalTransformation #DataAnalytics #AIChallenges #RealWorldAI #TechFailures #StrategicIntegration #Adaptability #EnterpriseLearning #BMJResearch #AIandHealth #ArtificialIntelligence