🤖 Day 12: Evaluate whether you trust AI to support testing and share your thoughts

[crmueller (Claudia Mueller)]

AI can help a tester providing test case and test data creation ideas.
I would never blindly trust anything an AI produces but validate its context.
(At the end of the day, I am responsible for my work and can’t blame an AI.)

What I would use AI for:
. small functional test cases
. extensive test data coverage
. performance testing
. repetitive steps

What I would not use AI for:
. integration testing
. E2E testing
. usage of sensitive data
. system security

As testers, we have to design & develop models and monitor them, collect & prepare good data, always check with legal/ethics teams.

[mrukavina (Marijana Rukavina)]

I work as a tester of ecommerce platforms. I worked in that domain for all of my professional career as a tester.

These articles mentioned in the assignment refer mostly to risks that are in general present when we speak about AI as a very broad, vague term and applied in different aspects of various disciplines, from science, education, environment, traffic, human resources, future recruitment, etc.

I have a feeling that risks mentioned there are on much global/higher level then it might apply to my current testing context.
But yet there are some I can connect to the ecommerce field, like:

1. Privacy concerns - how do we handle PI information and other sensitive data. I see big risk there. Even though a lot of regulation is there in place in comparison to the age before GDPR and things like that, it is still hard thing to put those regulation in practice and verify and identify the violation of the same.
2. Security risks - more AI power means also more tools for the attackers and the ways that they can steal and manipulate the data the the system holds. Or do harm in attacking the business and taking down the site, making our daily business stop and work against us.
3. Bias - favorizing certain product recommendations over others based on shady rules
4. Concentration of power - large companies dominating the market since they have the most resources to put in facilitating the AI.
5. Dependance on AI - sometimes I get “panic attacks” if chatGPT is not available since I “forgot” how to write a user story, test case or manipulate the data without it being my sidekick. Almost like parking a car and getting used to the sensors, which is great and helpful but if they stop working, I am in trouble. That one scares me a lot.
6. Loss of human connection - I still prefer to have a human mentor and collaborator, but AI option seems pretty convenient. Potential risk of alienating yourself from human touch.
7. Misinformation - I agree with @lisacrispin , if you don’t know anything about the topic you are looking at, you are in the danger zone. Simply because you don’t know. One should be crtical towards the outputs from AI and always double check it → my only advice is to think twice

Hah, at the end I managed to list quite of them, so maybe they are present in my day-to-day work level. More than I thought before dealing with this assignment …

[ayamfouad (Ayah Fouad)]

I skimmed through the provided articles, 10 Challenges Of Implementing AI In Quality Assurance, The Role Of AI In The Changing Landscape Of QA, and all your replies.

Here are my insights:

@billmatthews

• What context do you work in?

SaaS and E-commerce

@billmatthews

• What AI risks are introduced or amplified by the introduction of AI in Testing for your context?

Lack of human perspective/judgement, security, and privacy

@billmatthews

• Where should AI not be used in your testing context?

tbh, I would only use it to help with repetitive or predictable tasks

@billmatthews

• To what extent should the use of AI be trusted in your context?

I don’t trust tools generally, so I am putting 40% - it’s definitely helpful in accelerating some of the tedious tasks or with automation but would not rely on it for end-to-end scenarios for this context

@billmatthews

• How might trust for AI in Testing be increased in your context?

Not sure, think will keep interacting with it and observing changes to be able to determine this

[poojitha-chandra (Poojitha Manam)]

@billmatthews

Biased and discriminatory behaviours

An interesting article/story:

LLMs become more covertly racist with human intervention

[nao (Nao Deng)]

My Day 12 Task

1. About Research on AI Risks

I quickly read through the two recommended articles and summarized their key points:

Summary of the article [The 15 Biggest Risks of Artificial Intelligence]:

Artificial intelligence poses significant dangers and ethical challenges.

• Lack of Transparency: Complex AI decisions may lead to distrust.

• Bias and Discrimination: AI may perpetuate societal biases.

• Privacy Issues: AI can collect personal data, leading to privacy concerns.

• Security Risks: AI can be used for cyberattacks and autonomous weapons.

Summary of the article [Challenges of AI]:

Artificial intelligence carries potential benefits and risks but lacks unified regulation.

• Definition of AI: AI is defined as technology that performs tasks requiring human intelligence.

• Risks and Benefits of AI: It offers enormous potential advantages but also poses ethical, security, and societal risks.

• Regulation of AI: There’s a lack of unified regulation due to private sector dominance and government catching up.

• Ethical Issues with AI: Identifying and mitigating moral risks in design and ongoing usage is crucial.

Personal Thoughts

In general, from the theoretical proposal of AI to the implementation of related models and tools, there have always been unclear ethical dilemmas, inadequate regulation, and insecure data privacy. The risks of AI persist and, personally, I believe they won’t disappear.

Both articles address these points. Although AI is believed to be the future, many people still question the accuracy, data security, and fairness of results while using it. After all, the companies behind the operation of these AI tools face pressure from both governments and revenue.

2. About Reflection on the Role of AI in Testing and Sharing Your Insights

I believe there are risks associated with AI’s role in responding to testing-related results:

• The risk of ethical bias will undoubtedly affect the integrity of AI-generated testing data and scenarios. A biased AI may intentionally discard results that should be included.

• Data privacy and security risks make me cautious when interacting with AI, as I refrain from providing real contexts to prevent data collection. In our industry of internet software development, leaking data during the early stages of product release poses significant risks.

To mitigate these risks:

• Regarding ethical bias: My habit has always been to not entirely rely on or trust AI results. Instead, I use AI results to expand my thinking and generally perform a secondary human review of AI-generated testing data and scenarios to confirm their usability.

• Regarding data privacy risk: I apply partial obfuscation to the prompts and contexts when interacting with AI, reducing the exposure of real project and business information.

As I work in developing new internet products for clients, data privacy and security have always been red-line issues. Therefore, I am cautious when using AI in projects, and I use it to assist in repetitive or predictable tasks under the premise of avoiding risks.

My trust in AI results depends on the certainty of my current requirements. If my requirements are clear enough, I use AI more for time-saving and efficiency purposes, and I fully trust the results.

By using different AI tools for daily testing tasks and then manually judging the AI-generated responses, trust in the testing capabilities of certain AI tools is gradually enhanced.

My blog post : 30 Days of AI in Testing Challenge: Day 12: Evaluate whether you trust AI to support testing and share your thoughts | Nao's Blog

[schuur.e (Evelien van der Schuur)]

hello,
I don’t have a lot of time.
But the risks I see are:

1. Security risks
2. Building on the previous failures
3. You don’t know it’s correct

Role of AI in testing:
It can help to make creating test faster. But the risks is that you will allways make the same tests. Also you don’t know if security is up to date.

My insights:
I work in a financial company. And they are very scared of security risks. They have a lot of code tracking tools for vulnerabilties. You can’t put your code directly to an AI. You can’t share data with AI etc. We only use chatGPT for translations. And then we will read the translation if it’s real;y correct. Sometimes we askes chatGPT for a piece of coding. For everything we will try with AI, we have to think about the risks and what we share. Mayvb we have to ask before we start if it’s allowed to use a certain tool.

[markcole]

I trust AI only to the degree that I trust other humans. I do agree that the #1 problem with AI is lack of transparency. It is sometimes not that easy to track down where the AI is getting its information. It doesn’t always tell you. It has told me that it does not even know where the information came from. That is a secret. Well if it is a secret, then it cannot be verified. Therefore I cannot trust it.

AI is only as good as the data that it was trained on, and that data may be biased. The same is true with newspapers and books and anything humans touch. We only know what we know and we always have a bias. We just need to factor in that AI is biased by the information on what it has been trained on or failed to be trained on. Both real intelligence an artificial intelligence has biases, so I found bias to be less of a trust fail then I thought it was originally. I trust AI in the same way I trust non-artificial intelligence - inspite of bias.

[grcyxie (Grace Icay)]

Working in the field of technology as a QA Engineer, there has been a lot of discussion over whether AI will take over the jobs in the QA industry. After reading more articles and predictions on the matter, it’s projected that Software Developers are actually more likely to lose their jobs in the face of code generating AI tools. However, SDET or QA Automation Engineers are also likely to face the same challenges that Software Developers are now facing,. As a QA Engineer myself, I think there’s a lot more analysis and planning involved in the QA process in the context of my organization so it’s a lot less likely for AI to “take over” in this scenario.

Of course there are ethical and bias concerns over using AI to generate test data as well, but following the guidelines provided by UNESCO and IEEE or the organization’s internal guidelines could help alleviate some of the issues found here.

It’s likely that privacy and confidentiality are the top concerns for using AI in my testing context, as confidential data can be used in the app that my organization is working on. Unless there is a clear way to distinguish that the data is not being used to teach the LLM and endpoints are managed accordingly, there shouldn’t be too much of privacy concerns for it. As for the security aspect, using encrypted algorithms, firewalls and a reputable cloud service provider to host the application with the LLM could alleviate some of the concerns.

Granted, as AI is still in its developmental stages, it’s still difficult to completely trust it with abandon. And as different LLMs are still going through refinement and reconfigurations everyday, it’s less likely for me to stick to only one AI tool and I will still likely evaluate more tools that would be a better fit for my organization rather than adopt one right away.

[indres (Indre Stasiuniene)]

Hi, everyone,

for this day challenge I read recommended article and also 12 Dangers of Artificial Intelligence (AI) | Built In, We know the risks of AI — here’s how we can mitigate them | World Economic Forum (weforum.org), AI’s Role in Software Testing: Revolution or Risk? (proleed.academy)
I think, that biggest risks related with AI are those:

Privacy Concerns

AI systems often collect personal data to customize user experiences or to help train the AI models we using. AI technologies often collect and analyze large amounts of personal data, raising issues related to data privacy and security.

Security Risks

Hackers and malicious actors can harness the power of AI to develop more advanced cyberattacks, bypass security measures, and exploit vulnerabilities in systems.

Operational Risk

Potential treat to loss of trade secrets due accidentally misuses, negative impact of embracing AI, misconduct of AI, when it works not as expected.

Dependency on Continuous Data Input

Data relevance, quality, and consistency must be maintained within the data so that AI-based testing procedures can operate smoothly and efficiently. If it is not, then the results will be inaccurate and unreliable.

Misalignment of AI Predictions with Real-world Scenarios

The test scenarios and predictions generated by the AI models may be suitable for controlled testing settings, but they may not accurately represent the complex nature of the actual world.

How to minimize risk of AI:

Deal with own AI tools inside the company and establish own organizational AI standards, training staff.
Develop legal regulations
Assimilate safe data handling practices.

[vishnuar]

Thanks for the insights.

[vishnuar]

Hi

Interesting topic

What context do you work in?
Healthcare domain, Testing PACS application

What AI risks are introduced or amplified by the introduction of AI in Testing for your context?
Data privacy and security, If it is a sensitive data, It should be anonymized.

Where should AI not be used in your testing context?
I won’t use AI output to decide something during the testing phase but its output can be considered as a reference to think further

To what extent should the use of AI be trusted in your context?
Same as above answer

How might trust for AI in Testing be increased in your context?
If i get more evidence on Data privacy and security measures

[futurelady25 (Amruta Bhokare)]

• What context do you work in?
  –IT for Airline Industry
• What AI risks are introduced or amplified by the introduction of AI in Testing for your context?
  – Complexity and Understanding : AI systems, especially those using deep learning, can be highly complex, making it challenging to understand how they reach conclusions.
  – Over-Reliance on AI : There’s a risk of teams becoming overly reliant on AI tools, potentially neglecting the need for human judgment and creativity in testing.
• Where should AI not be used in your testing context?

1. Critical Decision-Making: AI may not be suitable for making critical decisions where human oversight and intuition are crucial.
2. New or Unexplored Scenarios: Testing scenarios that are entirely new or uncommon might not have sufficient data for AI to provide accurate results.
3. Ethical Considerations: AI should not be used in testing where ethical implications could arise, such as in privacy-sensitive areas without proper safeguards.

• To what extent should the use of AI be trusted in your context?
  Continuous validation against real-world scenarios and manual testing can help ensure AI’s accuracy.
• How might trust for AI in Testing be increased in your context?
  Education and Training : Providing training to testers on how AI works, its limitations, and how to interpret its results can increase trust.
  Pilot Projects : Starting with smaller-scale pilot projects allows teams to familiarize themselves with AI tools and build confidence gradually.

[sargam (Sargam Panwar)]

Your comments are really helpful, Trust me or not but after I read the task, the next thing I do is look for your comment and views about the topic!

[sargam (Sargam Panwar)]

Hello Everyone!
The first thing that came to my mind was ‘security of the data’ or ‘how well to trust AI tools in testing methods without actually reviewing it by ourselves’.
I just explored the two attached blogs and it was amazing to read something new about AI in testing.

Impact and Risk: The impact of using AI would be beneficial, time efficient, labor efficient, and cost-efficient as well but relying totally on AI would be something I will not follow up on since ‘AI makes decisions using algorithms that either follow rules or, in the case of machine learning, review large quantities of data to identify and follow patterns.’ and if in this case, either the algorithms or the rules and simultaneously the patterns governed by AI has even a small mistake, it can cost us more than we thought we are going to save using AI.
Also in the Day 4 AMA video, Carlos shares an issue regarding the ‘Context’ of the data which was a point that needs to be taken care of while using AI in Testing.

Also, the blogs mentioned above share a couple of more things regarding:

1. ‘Security and Privacy of Data’ while using an AI tool. - To fully understand how much risk the tool will cause to the data security and the individual’s privacy.
2. The need for AI to have a large amount of data sets to predict the patterns can be limiting to some organizations or individuals.
3. AI and its usage should not be based on gender, racist things, and mythological values, AI is a technology and not something that creates differences between people.
4. Also environmental problems should be taken care of with the increased use of AI. An increase in technology should be there but not at the cost of nature.
5. AI tools that people use to increase their followers on Instagram should be immediately banned, AI should be the source of goodness in someone and should not be used as a bane.

• ** Insights:**

• What context do you work in? : Precise Context-based Prompt leads to ‘Precise Context-based’ output.

• What AI risks are introduced or amplified by the introduction of AI in Testing for your context?

• To what extent should the use of AI be trusted in your context? AI should be used but just to ease our work and not to be dependent on it for our work done.