AI - Organisation’s Must-Have.
Problems Designing and Implementing AI systems.
I’m self taught, Learning new technologies, Aiming to implement them benefitting clients plus getting paid, My challenge, Stay ahead of howling pack of ravenous wolves.
Experimenting.
Machine learning, Flourished in ‘90s, Changed its goal from achieving artificial intelligence to tackling solvable problems of practical nature.
Decided to build Machine Learning system around 2010, Now have scars from decade of learning experimenting and teaching AI.
Built my machine learning machine heavy on graphics processing, Went open-source, Anaconda framework, Python language with Keras Sci-kit learn and Tensorflow, Able to build and demonstrate AI concepts.
Applying AI — Learn by Doing.
Contacted decision makers, Clients and directors, Managers and fellow developers, Conducted tailored demonstrations, Helping them understand AI’s benefits and pitfalls in quest to use AI within their organisations.
Often asked to carry out small well paid pilot studies, Focused on analysing implementation risks.
AI Implementation Risks.
Shouldn’t be surprised, Initial studies showed implementation of corporate AI not really a problem, Although in my opinion, There would be little by way of short term returns on investment.
Over several studies found main issue was fitness of data for reliable AI analysis.
Car sale yard, Desire was to minimise holding time of cars by optimising demand and supply, Idea seemed ok, First step was devise way of describing stock consistent across all outlets.
First obstacle was color, How would you describe “white”? Seems obvious but found different variations between makes and models, Constructing data dictionary difficult and time consuming, Adding to woes colour perception changes under different showroom lighting schemes, Analysis showed salesperson often key determinant to successful sale, Next job please!
One more example, Demonstrate AI to predict student retention, Silly me thought good idea, Institutions had multiple data sources between courses ranging from teacher’s spreadsheets to centralised systems, Found usual problem data definitions lacked consistency for reliable institution wide AI analysis.
Final Thoughts.
AI not effective when implemented as “bolted on” to existing systems.
Generative AI, Issues around using published on-line material for training purposes without authors’ permission, Litigation started.
Education institutions, Students using generative AI without disclaimers becoming subject to charges of plagiarism, Loss of marks and sometimes failing their courses.
Change is inevitable, Personally see AI as valuable tool, But feel transition will involve exercising greater control by Governments of all persuasions with increasingly hostile reception.
Blessed be.
