A lot depends on the input data as well as the quality of the underlying inference engine. Issues with either will lead to trouble. One of the challenges with tools like ChatGPT is the input data is highly questionable which leads to hallucinations and other types of bad inferences.
Effective prompt engineering is another way to reduce the likelihood of bad results - by constraining the scope of the decision making, it is possible to get better quality.
Like any tool, AI will only be as good (or as bad) as the person using it. Therefore, I would advance the theory that it needs to be driven by us personally. Equally so, the more we use and become familiar with a tool, the more likely we are to trust and to be able to recognise when a result is not inline with the desired outcome.
As Kiron stated above, by constraining and refining the scope of the decision making required, quality will improve. Saving Changes...
Sergio Luis ConteHelping to create solutions for everyone| Worldwide based OrganizationsBuenos Aires, Argentina
No trust. In fact, is the same than any software programs people write because is a software program. Then, you have to test it. To work with IA you need to use statistics and you can calculate the grade of confidence for each IA entity you create. Always IA will have a probability of error. The error will be acceptable or not depends on the domain where you are using IA. For example, is not the same when you work on health than in other domains. Saving Changes...
Markus KopkoAI Enabler for Project & Program Mgmt | Founder PMotion.ai / The PM
AI Coach| PMotion.aiHamburg, Hamburg, Germany
Dear Mourad,
Drawing from over two decades of experience in this domain, I'd like to share some best practices for controlling the results and information provided by AI to ensure their reliability and alignment with organizational objectives:
Data Quality Management: The foundation of reliable AI output lies in the quality of input data. Implementing robust data governance and conducting regular audits are crucial to maintain data integrity and prevent errors.
Algorithm Transparency and Explainability: Opt for AI systems that are transparent and explainable. This enables us to understand the rationale behind AI decisions, ensuring accountability and facilitating informed decision-making.
Continuous Monitoring and Evaluation: Establishing KPIs specific to AI performance is essential. Regular monitoring helps in assessing the system's effectiveness and making necessary adjustments.
Bias Detection and Mitigation: AI systems can inadvertently propagate biases. Implementing checks for bias and fairness is vital, particularly in decision-making processes that impact resources and personnel.
Feedback Loops: Integrating feedback mechanisms allows the AI to learn and adapt to our organization's unique requirements, enhancing its effectiveness and accuracy over time.
User Training and Literacy: Ensuring that our team members are trained to understand, interpret, and use AI outputs correctly is fundamental. This empowers them to leverage AI tools effectively.
Compliance and Ethical Considerations: Adherence to legal and ethical standards, especially regarding data privacy and ethical AI use, is non-negotiable. Regular compliance checks are essential.
Collaboration with AI Experts: Partnering with AI experts and data scientists helps us navigate the complexities of AI systems and make informed decisions about their deployment and control.
Customization and Configuration: Tailoring AI systems to our organization’s specific needs enhances their relevance and effectiveness. This might involve configuring the AI to align with our strategic objectives.
Risk Management: Identifying and managing risks associated with AI deployment, including technology dependency, system failures, and cybersecurity threats, is critical.
In conclusion, while AI substantially enhances process efficiency and decision-making, it requires careful management and control. The above strategies have been instrumental in my experience, ensuring that AI is a beneficial and reliable tool in our project management arsenal.
I look forward to further discussions and insights on this topic.
Best regards,
Markus Saving Changes...
Nelisiwe KhumaloSenior Project/Programme Manager| South African Revenue Services(SARS)Pretoria, Gauteng, South Africa
AI can be manipulated to perform the desired outcomes. Like any other technology, the result will have to be tested to determine whether they are in fact what was anticipated. On this basis it essentially works on data input like an other program. Saving Changes...
Anton OosthuizenSenior Business Analyst / Project Manager| Self EmployedPretoria, Gauteng, South Africa
By controlling the input.
...
1 reply by Zohaib Qadir
Nov 19, 2023 2:57 PM
Zohaib Qadir
...
I Second This.
Saving Changes...
Zohaib QadirSystem Administrator Picture Archiving and Communication System (PACS)| Peshawar Institute of CardiologyPeshawar, Kpk, Pakistan
I have to disagree with Anton although I understand it in theory. Take all the input you can get. Control how you filter and process it.
...
1 reply by Anton Oosthuizen
Nov 20, 2023 3:10 AM
Anton Oosthuizen
...
I 100% agree that you need to filter and process the output but I always prefer to start at the beginning when trying to solve a problem. The concept of GIGO is something few pay attention to these days because we think that AI will detect that we are stupid and meant to say this instead of that. That is fatal because instead of now sitting with 2 options to analyze and process, you end up with 1001, of which 999 are just crap. So by deduction if you control the input you have solved 99% of your output issues.
Saving Changes...
Anton OosthuizenSenior Business Analyst / Project Manager| Self EmployedPretoria, Gauteng, South Africa
Nov 19, 2023 7:08 PM
Replying to Keith Novak
...
I have to disagree with Anton although I understand it in theory. Take all the input you can get. Control how you filter and process it.
I 100% agree that you need to filter and process the output but I always prefer to start at the beginning when trying to solve a problem. The concept of GIGO is something few pay attention to these days because we think that AI will detect that we are stupid and meant to say this instead of that. That is fatal because instead of now sitting with 2 options to analyze and process, you end up with 1001, of which 999 are just crap. So by deduction if you control the input you have solved 99% of your output issues. Saving Changes...
Thomas WalentaGlobal Project Economy ExpertHackenheim, Germany
How can we control the results and information given by ANYBODY?
There are techniques available:
- build your trust in them (by heuristics or by reputation/hear-say))
- build redundancies (ask someone else)
- establish checks and balances (procedures, regulations)
- establish risk mitigations Saving Changes...
