The global skills and competency framework for the digital world

#35 Artificial Intelligence (AI) - New skill or skills? change request accepted

The importance of AI needs to be considered and introduced.

See the SFIA 8 theme for artifical intelligence

From Australian Public Sector SFIA Forum:

Need to consider a new skill AI to reflect the increased use of AI in decision making wrt large data systems.

Current status of this request: accepted

What we decided

Include in review of artificial intelligence and machine learning skills for SFIA 8

Ian Seward (General Manager)
Sep 11, 2017 11:50 AM

This came out of the Australian Public Sector Forum with a focus being on AI related to automated decision making associated with large data such as facial recognition or identity confirmation.

There is perhaps another angle to this in the area of autonomous machines. Need to further discuss scope of skill sin AI.

Matthew Burrows
Oct 18, 2017 07:51 PM

Had lots of discussions on AI and machine learning over the last year. At the moment only Analytics mentions this, but that doesn’t mean the other skill descriptions don’t apply. One discussion last week asked about programming AI applications and systems, and all agreed that PROG covered the coding aspects - it doesn’t make a specific reference to the purpose of the code, but if we included all the possible scenarios then the descriptions would be too long and require constant updating. The focus needs to be on describing what people actually do - “flying a plane” rather than “flying the following types of plane....”

Paul D Jagger
Dec 20, 2017 08:25 AM

Gentlemen, I for one welcome our new Computing overlords... More seriously the topic of AI is better expressed as Cognitive Computing, which is the new industrial revolution and combined with Robotics will be as significant a change to our lives as any major advance in technology has delivered. This is an opportunity for SFIA to be (for once) ahead of the game!

Ian Seward
Dec 20, 2017 09:44 AM

Great comment Paul. Yes I've spotted this as a 'theme' on a number of occasions hence the initial CR. I'm trying to find some people who might have some insights as to what might need to be in SFIA and have the debate any suggestions gratefully received.

As someone whose first job out of college was to research and develop high-survivability autonomous vehicles for work in hostile unstructured environments (it's all been a bit less fun since) I agree with Matthew some skills are still right e.g. PROG but what do the experts think are missing - let's have the input and discussion.

Matthew Burrows
Mar 05, 2020 10:45 AM

I thought this might be useful to help thinking as we start to think about SFIA8. I had a question from a customer yesterday, as follows:
"What skill would you use for someone who trains an artificial intelligence? As in a chat bot. Or a self learning document scanning system. Not really super-intelligent AIs, but still, self learning systems, that get trained by some people in our organisation. They don't code, they don't configure software, but they set up examples, decision trees, rules, guide the AI through masses of examples, then rinse, repeat, and start again, until the bot/AI is able to handle our specific product limitations, or the document scanning system can manage the written complaint or the random threatening letter."
BSMO was one of the skills they found useful, and also considered various options including INAN, TECH, BPRE, KNOW< DATM, METL, BPTS, DESN, SWDN, PROG, DTAN, INCA and HCEV.
This user is hoping that SFIA8 provides a better skill to describe this area.

Paul D Jagger
Mar 05, 2020 11:23 AM

Having just trained several ChatBots for various client environments the skills required to train a ChatBot must also include: Grammar and language skills in writing questions - the ChatBot must be able to express a users intent in at least 5, ideally 10+ utterances for each intent and that requires a set of skill not represented elsewhere in SFIA. For example "What's the best beginners course for AGILE", "Where should I start if I want to learn AGILE?", "How can I get started in AGILE learning?", "Where can I enrol on an AGILE foundation course?", "I want to start learning AGILE, where to I begin?"... all have the same intent but are 5 very different utterances.

If a user population includes users who are not native English speakers then the writing challenge becomes greater. Even for English speakers there are challenges across different nations, for example "How long does it take to sit the AGILE exam?" is "How long does it take to write the AGILE exam?" for Canadians and Americans... whereas in the UK we would consider 'writing' the exam as meaning 'authoring the content of the exam for others to sit' in USA/Canada it means 'the process of undertaking the exam as a candidate'.

Yeap ChatBots training certainly involves skills outside the norm of SFIA!

Carol Long
May 12, 2021 06:14 PM

I agree that we need to review SFIA but debates about English (and other language) syntax is akin to debating coding style. Can we unravel the root of these skills? Are they another form of technical analysis and encoding information (well within SFIA which is language independent) or something truly new?
There are some wider professional skills needed:
- some around privacy, data protection and data management/engineering, including safeguarding of data subjects/users (Peloton 2020 vulnerabilities flagged this) that are not within current scope
- system security has wider implications/responsibilities when data can predictable manipulate the actions of an autonomous system
- there is an area around logical transparency and decision explain-ability that looks new (they just used to be professional ethics, now an integral part of dev, test and system use)
- 'quality' reviews and evaluations probably need to expand into ethical reviews, and impacts on societal / social cohesion considerations and bias
- audits and compliance may need to review broader justice, equality, environmental standards because of the ways in which (particularly public sector) organisations are likely to use these systems
- governance needs to be clear about who is responsible for the actions of which autonomous systems and how boundaries will be resolved
- change control needs to assess what "control" it actually has for systems that could be setup to evolve in use, and how to manage the gaps
- risk management needs to look at intentional / unintentional harms for systems that have varying degrees of autonomy (beyond current practices that are quite constrained by boundaries, the future will be fuzzier about who is responsible unless there is proper action)

This is big topic for one CR!