Why is AI so challenging and what are the common pitfalls that lead to failure? In this blog post, Krystian Jabłoński, a senior AI advisor and researcher at Lingaro, will share his insights on why AI projects fail, including projects related to Generative AI. He will also share a few important guidelines for making AI projects succeed.
One of the main reasons why AI projects fail is unrealistic expectations. Many people have a vague or exaggerated idea of what AI can do, based on sci-fi movies or media hype. Take case of ChatGPT, for example. They think that AI can solve any problem, no matter how complex or ill-defined. They also expect AI to deliver quick and easy solutions, without considering the effort and resources required.
This leads to disappointment and frustration when AI projects encounter difficulties or limitations. AI is not a magic bullet that can fix everything. It is a probabilistic technology that requires careful planning, testing, and evaluation. AI projects also need clear and measurable goals, as well as realistic timelines and budgets.
Another reason why AI projects fail is choosing the wrong use case or not defining it properly. A use case is a specific problem or opportunity that AI can address, such as improving customer service, optimizing inventory, or detecting fraud. A good use case should have a clear business value, a high error tolerance, and a level of complexity that internal development teams could handle.
However, many AI projects fail to meet these criteria. They either choose use cases that have no real benefit or use cases that are too hard or risky to solve with AI. For example, some use cases may require very high accuracy or reliability, which AI might not be able to achieve. Some use cases may also involve ethical, legal, or social issues that AI might not be able to handle.
To avoid use case related issues, AI projects should start with a thorough analysis of the problem and the potential solution. They should also establish the success criteria and the performance metrics for the AI system, as well as the data and resources needed.
A third reason why AI projects fail is organizational constraints. These are the internal and external factors that affect the implementation and adoption of AI, such as budget, regulations, culture, and leadership. Many AI projects face challenges or obstacles due to these factors, which can slow down or derail the progress.
Difficulties in getting teams to collaborate could also block the progress of AI projects. Collaboration, by the way, is something that current AI models could not engage in.
For example, some AI projects may not have enough forecasted budget to support the development and maintenance of the AI system. These might also face regulatory hurdles or compliance issues, especially in sensitive domains like healthcare or finance. Some AI projects may also encounter resistance or skepticism from the stakeholders or the end-users, who might not trust or simply understand the AI system.
To overcome organizational constraints, AI projects need to have a strong business case and a clear vision for AI. They also need to have the support and sponsorship of the senior management, who can provide the necessary resources and guidance. Moreover, they need to communicate and collaborate with the relevant parties, such as the IT department, the legal team, and the customers, to ensure alignment and acceptance.
A fourth reason why AI projects fail is lack of key resources. These are the essential ingredients that enable the development and operation of the AI system, such as data and talent. Many AI projects struggle or fail because they do not have access to these resources in sufficient quality.
Data is the fuel of AI. Without enough correctly labelled data, AI cannot learn or perform well. Sadly, many AI projects face data-related challenges, such as data scarcity, low data quality, insufficient data labeling, weak data security, and poor data governance. These challenges can affect the accuracy, reliability, and scalability of the AI system.
Talent is the brain of AI. Without skilled and experienced people, AI cannot be built, managed and supported. However, many AI projects face talent-related challenges, such as talent shortage, talent gap, talent retention, and talent collaboration. These challenges can affect the innovation, efficiency, and sustainability of the AI system.
To address the lack of key resources, AI projects need to invest in and optimize their data and talent assets. They also need to leverage external partners and platforms, such as Lingaro, who can provide the expertise, solutions, and support for AI.
The fifth and final reason why AI projects fail is due to the inherent nature of AI itself. AI models, especially GenAI and Large Language Models, are non-deterministic, meaning they do not produce the same output every time. Achieving this capability requires extensive training, during which the AI learns to identify patterns to make predictions or generate new data.
For example, in area of preventive maintenance, AI model can predict when factory equipment will need servicing to prevent malfunctions and costly repairs. It learns by analyzing past system failures, studying the events leading up to these failures, and identifying root causes. By understanding these patterns and sequences, AI can alert operators about potential malfunctions, helping to avoid breakdowns.
Similarly, in the creation of product description pages, Large Language Models can generate unique and engaging content for each product. The AI learns from vast amounts of existing product descriptions, identifying effective language patterns and key features that attract customers. By doing so, it can create descriptions that highlight the best aspects of each product, improving customer engagement and boosting sales. However, due to the non-deterministic nature of these models, the output may vary each time, requiring continuous refinement and supervision to ensure consistency and quality.
However, the nature of probabilistic training introduces uncertainty as to whether the AI will recognize the patterns you want it to recognize and actually make the forecasts and predictions you expect. For example, heat is a common cause of machine breakdowns, but malfunctions due to the accumulation of dust and trapped debris might be missed because dust and debris are hard to notice and even harder to recognize patterns for. This highlights the importance of continuously monitoring the model, testing its performance, and retraining it when necessary.
These actions help turn AI projects around and prevent failure. However, these tasks are challenging when the model exhibits highly unpredictable behavior, and its process for producing output isn’t clear or easily explained to humans.
Now that we have discussed some of the reasons why AI projects fail, how can we make them succeed? Here are some guidelines based on the best practices and lessons learned from successful AI projects:
AI projects can be challenging, but they can also be rewarding. By avoiding the common pitfalls and following the guidelines, you can increase your chances of success and reap the benefits of AI. If you need help or guidance with your AI projects, feel free to contact us at Lingaro. We are a global leader in data and analytics, and we can help you with your AI journey.