There are many myths regarding Artificial Intelligence (AI) and Machine Learning (ML). As with other new technologies some are untrue and some are not. The myths that are untrue often create noise impairing us from making the right decisions about this new technology. In this blog post we will try to myth bust some of the most common myths within AI.
We will go through five myths which we meet when talking with customers and peers. The answers to each of the myths will be short to make it possible to be familiar with these topics without reading everything about them. The show is on - lets get on with some myth busting!
Myth #1: We need big amounts of data to do AI
This myth suggests that you need hundreds, million or billions of data points or in disk size terms gigabytes or terabytes before you can utilize AI and ML algorithms.
We have helped customers with what is considered a small dataset. However, the data we had available was the right data to solve the challenge. Right data is often easily discoverable by domain experts e.g. the person that operates or do the maintenance on a machine.
An example is the dataset we used in Automating tasks in healthcare with machine learning where the dataset consist of as little as 569 observations. However, it is possible to develop a model with good scores that can help automate a time consuming task. Another example is Predicting Machinery Breakdown on a Water Pump where we use a small dataset only containing megabytes of data to do predictive maintenance.
It is true, that the model often gets a better prediction with more data. However, you can gain value with small datasets, and improve the model over time.
This myth is busted! It is possible to use AI within your company without big amounts of data as long as it is the right data.
Myth #2: AI is a magic black box that can solve my problem
We are currently seeing many different approaches to how AI is delivered to customers. Some offerings focus on a “Give us your data and we will solve your problem” approach. This has two major drawbacks: (1) You do not have a specific problem in mind and (2) it is a black box approach to AI as you do not have to do anything expect giving access to your data.
The challenge with not having a specific problem in mind is that data does not tell much without giving it a context. The results to this black box approach is average at best because without domain knowledge important information about the problem is often lost. Further, many of our customers have told us that they have tried such an approach and that it have failed within their company.
The alternative approach is to to have a specific well defined problem to work with, e.g. predict when the bottleneck machine breaks down. In our experience this heightens the success rate ten folds for adopting AI within a company.
The answer to this myth is maybe but likely not. Doing AI within your company requires something from the company because it posses domain knowledge which is valuable when trying to solve a specific problem.
Myth #3: AI will take away jobs
This myth is one of the older myths within artificial intelligence. The media have for a long time focused on if AI will take away jobs in different industries.
What we currently see is that AI can automate tasks. Taking away one task will likely not remove a person's job but rather give them time to do something more important within the organization, something that requires a human, like making decision and using empathy. AI can be used as a collaborative tool to gain insight, and make decisions in a data-driven manner. But the insight is meant as a supplement for employees which they can use in their decisions.
The answer to this myth is likely not, but it will definitely redefine some jobs. Repetitive tasks will be moved away from people in order to free their time for more important tasks such as decision making.
Myth #4: Decision made by AI cannot be explained
This myth involves the belief that AI is inexplainable and it therefore cannot be used in many processes where transparency is important. An example of this is that the bank need to be able to say why you did or did not get accepted to have a loan. To investigate this myth we need to go a little deeper into machine learning and the different approaches within.
Quick Deep Dive into Machine Learning
In machine learning there are different approaches to solving a given task. Some methods give traceability and transparency from the start such as decision trees, Bayesian classifiers, and random forest. These are explainable because you can see how the model produced a given output be that a prediction or classification. An example using a decision tree can be seen in the figure below.
It is true that deep learning and neural networks are hard to reason about and sometimes it can be so hard that it can be almost impossible to know what lead to a given output. With neural networks it is possible to get information about how each connection in the network is weighted, which can give some information to why it came to the conclusion. There are currently being research done into explainable AI which tries to solve the transference and traceability problems within deep learning. Explainable AI (XAI) is a collection of methods and techniques which can make the results of the model understandable. However, it is still in a research state where it is hard to generalize and maybe it will stay this way for years or forever.
This myth is partially true. Depending on your challenge and the method used for solving it it can be more or less explainable. However, many of the challenges we help customers solve use methods which can be explained.
Myth #5: AI is only for big companies
The myth goes that AI is expensive, hard to do, and only benefit big companies. This myth might rise due to the first myth you need big amounts of data. It is possible for small and medium enterprises to utilize AI for getting insight or a competitive advantage. Solutions utilizing AI does not have to be expensive nor take long time to develop. They can actually bring value quickly. The most important thing for a successful implementation is the people and mindset within the organization.
This myth is busted. Small and Medium Enterprises (SMEs) can gain the same benefits utilizing AI as large companies.
This concludes our first myth busting AI series. I hope it gives you more confidence in taking the right decision for your company based on facts and not myths!
// Rasmus Steniche, CEO @ neurospace