This step-by-step guide helps public sector organizations use machine learning and artificial intelligence. Critical to this is identifying the right opportunities that add value to the organization while using AI responsibly, and building cross-functional teams. We also examine the lifecycle of a typical machine learning solution, including product development, deployment, and monitoring in order to drive continuous improvements.
Learn more: https://go.aws/49CSDqs
Speakers:
Marion Eigner
AI Strategist, AWS Generative AI Innovation Center
Neil Mackin
Principal ML Strategist, AWS Generative AI Innovation Center
About AWS
Amazon Web Services (AWS) is the world’s most comprehensive and broadly adopted cloud platform, offering over 200 fully featured services from data centers globally. Millions of customers—including the fastest-growing startups, largest enterprises, and leading government agencies—are using AWS to lower costs, become more agile, and innovate faster.
AWS in the Public Sector paves the way for innovation and supports world-changing projects in government, education, nonprofit, and healthcare organizations. Check out: https://go.aws/2XV2QeV.
Subscribe to our channel: https://bit.ly/2PQ1wIW
Watch our video catalog: https://bit.ly/38F5Zn1
#publicsector #genAI #generativeartificialintelligence #machinelearning #ML #AWS
Welcome back to the AWS Institute masterclass on artificial intelligence in this module we’ll discuss how to get started with machine learning and artificial intelligence including generative Ai and it all starts with the identification of an opportunity an opportunity that adds high value to your organization and viably and responsibly
Be realized using machine learning Technologies to identify such opportunity it’s a good practice to form a cross functional team consisting of a subject matter experts that knows your organization and their processes and is able to spot opportunities that lead to high value a subject mattera expert that can judge the technical feasibility of
Machine learning Technologies and is aware of their capabilities as well as a limitation just one example large language models are able to generate plausible sentences but do not have a real understanding of the world and lastly a subject matter expert who can facilitate the overall Discovery process I think you’re so right actually
On this need for cross functional teams you I’ve seen it I’ve seen actually go wrong when people haven’t done that which was so disappointing all the energy they put in thinking about a bank actually where they they had a load of technical experts building out a really interesting technical ml solution to
Categorize transactions and build a model of the behavior but they never talk to their line of business teams and so they never actually managed to get it into production because without the line of business teams you’re never going to find the right use case to use this so
It was kind of stalled and styed which is a shame after all that work um also I’ve seen it go the other way where business teams kind of go too far without bringing in the data experts or the technical experts in one particular case where they went a really long way
Down scoping this use case and quantifying everything we could do and then discovered when they talk to the data people that actually the data they thought they could get hold of would be really difficult perhaps it was managed by a third party supplier they’d need to have contract changes to make that data
Available there’d be additional costs that actually torped the entire project yes you’re so right I I’ve seen this various times and actually building such a cross functional team at the start can really help to create a backlog of opportunities that support you and your general leadership to Envision the
Overall opportunity that you can tap into using machine learning within your organization but as well help you to prioritize the different opportunities against each other and doing this priorization you can define a set of different factors that cover technical feasibility helping you to judge if a machine learning model and
Solution can be built to help you decide if it’s worth and desired building and a factor to judge on viability for your organization helping you to judge if you can viably operate that Solution on the long term so coming back to the examples you just shared you want to make sure at
The very beginning that you have the data available to realize an opportunity as well as that a business process that might be improved also can be approved and adapted towards a new uh method or a new flow to Foster responsible use of artificial intelligence is recommended to include diverse backgrounds
Perspectives skills and experiences in the priorization to reflect the overall impact and consider requirements for security privacy explainability auditability and legal compliance once you’ve done this priorization you should discuss the results with your executive leadership team in order to drive alignment and form a first understanding of what capabilities must be built to
Develop such a machine learning model and to secure the required resources for the development of a first pilot so so far we’ve talked quite a bit about the people and the organization side which is really important but perhaps it would be helpful if we think about machine learning from the perspective of the
Life cycle or the stages that you have in a machine learning model life cycle life stage I actually worked on ad’s well architectured framework ml lens which is a a document available on on the internet and we use there this diagram that actually illustrates this so as you talked around identifying the
Opportunity that maps on to the business goal how do you identify the first business goal characterize the business value and be very crisp and precise about that the next stage is really how do you then formulate that into a problem taking uh an analytical framing for What would do with data and
Algorithms to actually tackle that problem thereafter the side of data Gathering the data that you need perhaps transforming the data creating features in the data that the machine learning can grp on to in using the algorithm to create the models then the model development itself which approaches will use which
Algorithms will use or would you use generative Ai and if so which Foundation models would you use and how would you actually go about using them and then moving on to the area of deployment so taking what you’ve built and you’re experimented each world and deploying it into production so that it
Can be used by everybody before monitoring it and actually capturing how accurate is it how well is it being used how quick how performant how expensive how much value is it creating for us and each of these needs to go in a big circle we actually drive kind of a
Continuous Improvement cycle as we go around this Loop so you can see from the monitoring it might inspire you to do more business problems or it might inspire you to go back and rebuild the model to be more accurate in fact although it’s a single cycle you can
Often see kind of little Loops within it as you go back from trying to build out a model and thinking maybe if we had more data maybe if we pulled in more data in different feeds we could build out better features and build a better model or maybe monitoring the accuracy
Discovering is not actually quite good enough and going back to that model development stage to build a better model that’s more accurate and more effective so you could go round and round these mini Loops as well as round and around the overall Loop to drive that continuous Improvement we need so
This Loop will show you that operating machine learning models substantially differ to operating traditional software so it requires leaders to transform the organization so that they can operate machine learning models and up until now these machine learning models were mostly trained on your very own data and
You were the owner of the resulting machine learning model now with the availability of foundation models you can also use third party model that are pre-trained on vast amount of data that do not belong to your organization this can be a huge accelerator but also introduces additional dependencies and
Responsibilities let’s take an example imagine you want to build a conversational interface like a chatbot that uses user feedback to continuously improve over time by doing so the data from your user becomes an integral part of the model development cycle so it’s your responsibility to collect this data
Responsibly assure that you can use it in the intended need and that you reduce the bias that is or might be in the data before introducing it into the back into the model development life cycle you’re absolutely right there’s so much care need in this area I wonder if
We could think about you a bigger example you know Swindon bur council is a local government organization in England and they had a they had an interesting example which is about fly tipping which is not a very common sort of thing to think about initially when you think about artificial intelligence
Or machine learning fly tipping is actually where folks dump their waste where they shouldn’t so instead of like you or I would getting the council to come and collect it or taking it to the waste in immunity site they just go and throw it in some lane or back alley and
Leave it for someone else to deal with and that’s somebody else dealing with it usually involves one of the other residents ringing the council and the council to come out and collect it so they wanted to use AR intelligence in this area to see if it could be improved
Could they make it quicker for citizens to report it could they make it more effective so that AI could actually detect what has been dumped maybe critically figure out what kind of size vehicle we need to bring you know is it going to be a sofa that needs a large
Pickup truck or is it something simpler or actually is it toxic waste is it maybe related paraphernalia from drug abuse where we need to go and collect it quickly and try and dispose of it before it causes was even more harm so there a number of different challenges there
They put together a cross functional team as we were talking they brought together the customer service people with the digital team who built out the app with the waste collection people with also the data and it people and they joined all these together to think about how they could tackle this you
Think about it’s actually quite a cross functional application it transforms different parts of the business so it’s going to change the way citizens engage with the council and Report it’s going to change the customer service workload it’s going to change the way the waste management team behave and how they’re
Going to respond to to different incidents perhaps even change the rostering and The Roots it’s also going to change the it in terms of the data the data capture records management all of these put together is quite a bit they tackled this project and actually the whole project took three months from
The idea through to the deployment really quite quick and really effective and they did this with a delivery of just four people four people in this team and now Swindon are actually saving £3,000 a year on the fuel cost alone and obviously the environmental benefits as well carbon not being wasted but more
Than that it’s actually saving the council over 2,000 hours every year in terms of the cost of the team to do this and for the citizens it’s improved for them as well the average cleanup rate from an incident being reported to it being dealt with has dropped from 10
Days to in the four massive Improvement that’s a really great example and I hope that this one as well as the others really inspired you to spot some potential opportunities in your organization already and we can soon congratulate you on the achievement of your very first Milestone the identification and qualification of your
First machine learning opportunities as well as the formation of your first cross functional team next we’ll explore how to validate the technical feasibility of your machine learning solution and building a first productive solution so see you in the next module