Top 4 reasons Why AI Projects fail?
For those who are new to AI technology, Artificial Intelligence is intelligence demonstrated by machines, in contrast to the natural intelligence displayed by humans. The term “artificial intelligence” is often used to describe machines that mimic “cognitive” functions that humans associate with the human mind, such as “learning” and “problem-solving”. To understand more on AI read this blog article from Tejas
Despite increased interest and adoption of artificial intelligence (AI) in the enterprise, 85% of AI projects ultimately fail to deliver on their intended promises to business, according to a recent industry report. In this blog post, I have put together the top 4 reasons for failures
1. Organizing your AI team
Building your own AI team is not at all easy. Unless you have a budget the size of Google’s, building an in-house data science team is costly and time-consuming. There is a shortage of data science experts everywhere, and the experienced ones are often involved in engaging projects with big projects and companies. So, in this constrained environment, what is the best way forward? If you are starting fresh then trying to build a big team upfront is absolutely unnecessary. At the same time, starting out with college graduates and training them to become experts can be a costly mistake. In the quest of saving dollars, you are compromising on the quality of the deliverable. Experienced folks, usually from engineering background, with deep understanding of the business and the data are the right choice to start with. In AskSid.ai we started without an AI expert in the team. We put our best and experienced folks to take a dig at our AI expeditions. We laid out an implementation plan with well-defined and small milestones. One needs to get into a hypothesis and experimentation mind-set. Validate ideas quicker, eliminate or accept based on experimentation outcome and move ahead. In this process, we learnt from our mistakes and we improved at every step and came back strong. The first set of AI engineers then mentored the fresh and lesser experienced AI folks and helped them in productionizing their AI experiments. Today we have a sizeable AI experts team who can go all guns given a problem ahead of them.
2. Lack of proper Data Strategy
The importance of data strategy is often underestimated. In the past, data was just a by-product of processes and business activities. It’s different now. Data is the most valuable resource that allows companies to gain competitive advantage and come up with new ways of improving their operations. In most cases when dealing with AI projects, collection of data is a challenge. There’s not enough of it and if you are somehow able to get hold of the data then there can be issues with labelling, training data, etc. Data doesn’t just have to exist, it has to be labelled. If data isn’t properly organized, humans have to devote their time to the tedious task of labelling it. Data labelling is troublesome and time consuming, yet somehow many companies just don’t think about it at all. Because an AI system can only be as good as the data it’s fed with, you can’t have any tangible results if there’s no meaningful data behind it. Here at AskSid.ai, we have tools developed to automate data collection, validate data quality and easily manage it. You have to treat your data like gold and ensure it is not corrupted by any means and over the last 3 years, we have tried to carefully automate a large part of this process.
3. Unrealistic expectations from AI technology
Many organizations are tempted to follow the big players like Amazon, Google, Microsoft, etc. It’s just logical to assume that what they do makes sense. However, companies like Amazon have gathered enough capital to test various, even seemingly crazy solutions with very lofty goals. They can afford it, and experimenting with technology is great. However, smaller companies cannot afford to take so much risk. Observing the giants, learning from their successes and failures – that does bring value. But keep in mind that each business is different and has different needs. Don’t let yourself lose sight of what matters to you. Do some research, get your strategy ready, and then play your game accordingly. Implementing a new piece of technology can be challenging, and that’s something organizations must always remember. AI is not a magical tool – it needs the right use case matching your objectives, the right data, and success criteria, it also has certain limitations. When you identify the AI usecases you think is right for you, learn both the possibilities and the limitations the different AI solutions have to offer. It’s an important part of the process to know what AI can do for your business, realistically, and be able to later track the performance of the deployed models. Please remember – there is no such thing as perfection, even for AI. AI also makes mistakes. No AI algorithms will be 100% error-free. In our case, we have a Question and Answer system which can answer user questions asked through the Chatbot on the client’s ecommerce brand website. There will be times when the AI model is less confident about which question the user is asking and therefore what should be the answer. This is the case when AI is not able to clearly decide from its trained AI algorithms. In such cases, we handle this gracefully by providing the user with Did you mean? – Question suggestions. Small strategies like these will help your organization avoid your AI failing badly but more importantly also allows the AI to self-learn very quickly. Continuing on the same example, today our AI technology is able to answer 80% of all questions asked by website visitors of our client thereby prompting the client to take our solution and launch it even in their retail stores.
4. AI Bias
AI bias, or algorithmic bias, describes systematic and repeatable errors in a computer system that create unfair outcomes. Though the name suggests that AI is at fault, as described above, it really is all about people and how the AI is trained. Chief Decision Scientist at Google, writes: “No technology is free of its creators. Despite our fondest sci-fi wishes, there’s no such thing as ML/AI systems that are truly separate and autonomous…because they start with us. All technology is an echo of the wishes of whoever built it.”.
We faced this situation in AskSid.ai with our ecommerce NLP intent model. We saw that more and more user questions and complaints were classified as product discovery. What this means is questions like “How to wash cotton tops?” showed cotton top products, which clearly is not what the user wanted. Upon investigating we got to know that the engineer who tagged it was more biased towards classifying such sentences as product discovery. This created an imbalance in the knowledge-graph which lead to incorrect and biased intents model. To solve this, we had to revisit the labelling and correct it. To avoid such problems, it is advisable that more and more people validate the labelling before it is pushed to the LIVE environment. Again, this does not mean extensive manual effort but smart algorithmic ways to sample the right data set and validate is enough and can do the trick.
One famous example of AI Bias is Microsoft’s AI Chatbot Tay – which could automatically reply to people and engage in “casual and playful conversation” on Twitter. By flooding the Chatbot with a deluge of racist, misogynistic, and anti-Semitic tweets, Twitter users turned Tay – a Chatbot that was described as “a robot parrot with an internet connection” – into a mouthpiece for a terrifying ideology. After a failed effort to clean up Tay’s timeline, Microsoft pulled the plug on their unfortunate AI chatbot experiment.
Conclusion: So, how can you ensure a successful AI implementation?
The good thing is that with so many organizations having already failed at AI, you can learn from their mistakes and avoid making the same ones. It’s a good practice to observe the market, not just in your direct competition, but also in the tech world. This way, you will know what you can realistically expect, what use cases are promising, what limitations you have to take into consideration. Take ‘Fail Fast, Fail Early’ as a motto when it comes to AI. You have to experiment, test hypotheses, go down some blind alleys, learn from them and use these to find success. Having processes and methodologies for agile, rapid experimentation that delivers results quickly and at a relatively low cost is critical for success. Do not expect to draw up the perfect AI strategy for your business at the start and to execute it. Start with the assumption that you will have to evolve and change this as you progress.
If you like this blog or have anything to share, please leave your comments at firstname.lastname@example.org