Artificial intelligence sets the trends in all business areas, and as a tech company, we stay updated with the latest trends. We mainly follow those that positively impact productivity and performance. That is why the Artificial Intelligence tools were on our radar, and we were excited to test them.

AI-based solutions have the potential to greatly assist all of our team members. For instance, they enable developers to code faster, automate repetitive tasks, and concentrate on solving complex and challenging problems. We were curious how these marketing promises compare to reality.

Working with a new tool or technology is like embarking on a new mountain trail for the first time. It’s thrilling, and you want to reach the peak as soon as possible. But to get there safely, you must carefully plan your trip. You analyze the map, choose the correct equipment, and observe the weather. We applied the same principles to integrating Artificial Intelligence tools into our workflows. Our leaders conducted research and engaged the team in brainstorming sessions.

Man standing with umbrella lightning strike - a metaphore for the risk of use of artificial intelligence tools — An example of AI generated image reflecting on “Forewarned is forearmed”

Forewarned is forearmed

What questions do you ask yourself before going to the mountains? You usually check the weather forecast to see if there will be a storm. You review the specific trail details to decide whether to use shoe grips because they could be slippery. Focusing on the risks helps us prepare better for the journey.

We did the same: we gathered as much data as possible and then analyzed it. Our research on Artificial Intelligence flaws gave us valuable insights into a few areas that require our attention.

Privacy and security

Privacy is one of the first topics to be considered when searching for Artificial Intelligence drawbacks. Organizations that wish to utilize generative Artificial Intelligence should prioritize safeguarding data privacy.

Conversations with Artificial Intelligence models are usually not entirely private, even though not every tool will use them as training data. We should be cautious and avoid sharing confidential or sensitive information. It’s essential to anonymize any examples used and not unquestioningly trust technology.

If you meet a stranger during your mountain trip, don’t tell them your family secrets.

Clear accountability

Once someone starts using Artificial Intelligence assistants, they might feel less responsible for the results of their work. However, understanding the model’s limitations will help dispel this false sense of security and trust. For instance, the model’s training data is usually from the past; sometimes, it is from the last quarter or even half a year ago.

Additionally, Artificial Intelligence systems can be overly confident and generate responses that appear valid but are not true, a phenomenon known as hallucinations. The model’s primary goal is to provide as much content as possible. This is why the human factor remains crucial in conversations between people and machines.

Humans are responsible for supervising Artificial Intelligence recommendations and ensuring no errors are made. We control what part of the AI-generated data will be used. Moreover, we can guide the model in the correct direction, for example, by asking additional questions or pointing out mistakes (it’s called reflection).

Code quality and safety

Studies suggest that code generated by Artificial Intelligence coding assistants may be less secure than code written by a programmer from scratch. Moreover, the general code quality will vary: the recommended solution is not always the best and the most optimal. The training data wasn’t carefully curated; open-source projects or apps also included errors or bad practices.

Sometimes, assistants propose using dependencies that do not exist or, in the worst case, are insecure. Unique kinds of attacks pollute model data. There is no such thing as an entirely trustworthy Artificial Intelligence.

During your mountain hike, you will meet many different people. They can provide various advice and, most often, in good faith. Experienced hikers know that they have to review them carefully. If someone made a mistake and showed you an incorrect path, it will be your problem that you got lost.

Code complexity

Less experienced developers will feel tempted to use Artificial Intelligence systems to generate larger blocks of code for them. However, it is dangerous if the developer doesn’t fully understand the code written by generative Artificial Intelligence coding assistants.

It’s like preparing for a trip to the mountains with gear you’re not familiar with. If you don’t know how your equipment works, you won’t be sure it’s reliable or suited for the challenges ahead. Furthermore, if there are any problems later on, you might not know how to resolve them.

From the business perspective, easiness of maintenance and extendability are essential parts of the application design. Achieving these goals is difficult when the code is complex and challenging to read.

Knowledge of artificial intelligence makes the difference in security perception. — An example of AI generated image reflecting on “Knowledge is power”.

Knowledge is power

We had concrete goals for our discovery: asses the risks and examine the organization’s requirements for Artificial Intelligence systems.

Our extended research provided valuable insights that allowed us to search for the tools that could provide the most value for us and where the risks could be easily lowered. We knew checking privacy policies, terms of use, and general safety was essential. Settings that can be shared within the company were additional advantages.

General principles & ethics

Artificial Intelligence tools may initially seem like magic, with machines answering our questions elegantly and resolving tasks swiftly. However, it is essential to recognize that Artificial Intelligence is enhancing human capabilities, not replacing them. Human oversight is crucial in reviewing results and making final decisions.

Training

Workshops are a great way to share knowledge and learn how to use the tools efficiently. We’ve arranged external training for developers and will host internal workshops for other team members, focusing on practical application and security measures. Practicing our skills is a key.

Picture yourself heading to the mountains unprepared. You may reach the top, but you’ll be worn out and likely injured.

Best practices & policies

Apart from training, creating organization-wide policies and instructions is essential. If someone has a question, they know that there are multiple resources where they can find the answer.

Our policies include approved tools with usage limitations and recommended settings. Furthermore, we focused on making them both clear and straightforward to apply. Our shared knowledge base is constantly updated, and all team members participate in this process.

Key principles

We decided to emphasize these fundamental principles:

never share confidential or sensitive information,
never accept the Artificial Intelligence response without reviewing it,
never accept a response you don’t understand or cannot verify.

The policies are like a map or compass. Using them will lead you directly to the summit, while you might get lost without them.

Additional tips for developers

Apart from general principles, we also identified areas that require special attention from developers who use Artificial Intelligence coding assistants. We have already used a code review as a great tool to improve code quality and better understand the code.

Detailed and thorough code reviews also help lower the risks of using Artificial Intelligence tools. Another developer may spot the errors, propose better optimizations, and ensure that newly added code is consistent with the project.

We also use automated static code analysis tools to help identify potential vulnerabilities and security issues. Our programmers understand that they are accountable for their code. Despite receiving guidance from Artificial Intelligence, coding is their specialty, and they always verify and test the solution themselves.

Ongoing monitoring

With tools evolving so rapidly, such as those that use Artificial Intelligence solutions, it is crucial to review our tools and policies constantly. Are there any new security or privacy concerns? Have any new regulations been introduced? Is there a new tool that would work even better than our current one?

Tech companies are used to a rapidly changing world. We understand that our policies will be constantly updated, and some of the recommendations we give our team members will change. We continuously monitor the situation and gather new data. Once we make a change, we are transparent about it, ensuring everyone is aware.

Summary

We strongly believe that shipping reliable and secure applications is vital. When identifying opportunities to enhance productivity and performance, we seek the most effective ways to integrate these tools into our workflows. Understanding the risks and vulnerabilities of new technology enables us to do so responsibly. Moreover, involving human experts in the process yields the best outcomes.

Our backpacks are packed, and our shoes are laced up. We’re all set for this adventure.

If your project requires an Artificial Intelligence tool or you are planning to use AI-based assistance and are afraid of using it, don’t hesitate to consult us before proceeding.

Agnieszka Pastuszczak

10 Tips for a Responsible use of Artificial Intelligence Tools