Artificial Intelligence – Four aspects to consider when implementing AI

When implementing AI in your organisation, there are at least 4 important areas you should focus on.

Artificial intelligence is more than just a buzzword. With the explosion of data and the corresponding increase in the data storage capacity for storage devices, there is a pressing need to be able to make automated informed decisions based on available facts. Due to advances in technology as well as an increase in computing processing power, powerful computer software containing AI algorithms can now be developed to take advantage of the vast amount of data available and hopefully do something useful with it – apply the knowledge in a specific field to solve a particular problem.

Think of the human brain, the human brain itself has around 100 billion neurons. The average human is trained in a certain field over a number of years and throughout their lifetime, they gradually develop the neuron connections in their brain through experience, trial & error, learning at an institution and other learning methods. For example, a human accountant would have to have had some training, gained some experience, stored the experiences via memory encoding in the neurons in their brain but for them to be effective at their workplace they need to be able to consistently remember or retrieve the knowledge they have gained, apply the knowledge correctly and it is essential that they also continue to update or learn new things (i.e. regulations, new guidelines etc).

In a very similar way, artificial intelligence generally focuses on mimicking the development of the human brain by developing artificial neurons with interconnecting nodes. Just like when an average human is trained, a Research and Development team most likely also train an AI using training data that involves a combination of experience, trial and error, pure research from reputable sources and other learning methods. This in a sense, helps the AI to be able to determine what is right and what is wrong. When subsequently deployed in the field, the AI is then able to make automated decisions based on its perception, decision-making algorithms and the training it has received. GIGO (garbage in, garbage out) really applies here though, if the AI has been fed low-quality training data, then it will probably make low-quality decisions and it might not even be able to learn from the mistakes it has made. This is why a strategic quality review of the training data is of absolute importance. We now briefly consider 4 aspects to focus on when implementing AI. Focus on –

1. Having the Right Ingredients: Within many companies, it is the responsibility of the technical lead to ensure that the right architecture, research team, programming languages, methodology, product roadmap, algorithms, training data, design blueprints have been carefully selected; they should also have been developed the right way and at the right time. There might be other departments that actually govern or run the overall project. An agile development methodology allows incremental changes to be made and released so that the AI is exposed to real-world field conditions; this methodology is very useful when an AI is deployed in a sector where a certain percentage of mistakes below a certain failure threshold (e.g. 5-10%) is acceptable. The end-users are also able to immediately enjoy the benefits of the AI as opposed to not having anything at all to assist them or having to wait for years before they can enjoy the benefits of the AI.

2. Continuous Improvement: The key to an individual being the best person they can be is probably continuous improvement, constantly developing the right qualities and asking the right questions; there is also a need to acquire more knowledge and make better decisions. It is the same with AI. There is a need for the research & development team to recognise the importance of the need to start simple but to constantly improve and further develop the artificial brain of the AI (more knowledge, more training, more connections, more meaningful data, more tough questions, more testing, more quality reviews). It is critical to also remember the very important ethical side of AI.

3. The After-effects of AI: There is a need to carefully manage the after-effects of AI so that its impact is very positive for all parties. Even though many workers are worried about being replaced by machines, it might be possible to make the process a win-win for all parties. Important questions to consider are:

1. 1. 1. Have some workers been displaced by an AI?
      2. Are the end-users able to work successfully, comfortably with an AI – e.g. an AI-Assisted work pattern?
      3. In the long-term, what is being done to redeploy the employees whose jobs have been wiped out?
      4. Is it possible to utilise the knowledge the employees have by getting them to contribute their knowledge and experience to help the AI to grow?
      5. Do people really trust that the AI will serve them just as well?
      6. Can people objectively complain about the AI, if they are dissatisfied?

There might be a learning need to educate users about the after-effects of AI deployment and what the company is actively doing to monitor the situation to ensure that it is a very positive experience for all parties. The communication from senior management at regular intervals can highlight the benefits of the AI, positive experiences and what is honestly being done to reduce some of the negative experiences and perceptions. It might be a great idea to also encourage anyone to be able to freely ask questions about AI and its use, this really encourages transparency and accountability.

4. The Framework, Policies and Guidelines: There is also a lot of talk about explainable AI so that the AI’s decisions can be somehow traced back in a logical way. Even amongst experienced industry experts and technical analysts, there are so many differing opinions on what AI should do and what it should not do. There are also those that approach it from just the theoretical or whitepaper side, but many unfortunately find out that even though it works very well in theory, they soon discover that their pure AI theories do not really add up in practice when it is time to implement it in the code/architecture. Industries also differ and the way each company chooses to implement AI will always be slightly different. For instance, AI code in a speech-recognition module is significantly different from AI code for a hardware chip in a smartphone, AI fraud-detection code in FinTech or even AI code to support teachers in EdTech.

Responsible AI Development

There is a popular saying that – with great power comes great responsibility. The conclusion really is that there are so many aspects to consider when implementing artificial intelligence, remember that developing artificial intelligence comes with great responsibility. Will we ever get a generic comprehensive Artificial Intelligence Development Standard to benchmark our work against when developing an AI?