0:04 many businesses think of a large or a
0:07 small company say have many people doing
0:10 many different TS there's a framework
0:12 that had originated in economics due to
0:15 Eric Brunson Tom Mitchell and Daniel
0:18 Rock for analyzing the work task for
0:21 possible automation using Ai and this
0:23 framework turns out to be useful not
0:25 just for Economist to understand the
0:27 financial or economic impact of AI but
0:30 also for businesses identify specific
0:32 specific opportunities to use generative
0:35 AI let's take a look at how to do this
0:37 whereas there's been a lot of discussion
0:39 for example in the media about will AI
0:42 automate jobs it turns out that from a
0:44 technical and business perspective it's
0:46 more useful to think of AI not as
0:50 automating jobs but as automating tasks
0:52 and it turns out most jobs involve a
0:54 collection of many tasks let's look at
0:57 an example a customer service
0:59 representative will do a number of
1:01 different toss including maybe answer
1:03 inbound phone calls from customers
1:06 answer customer chat queries via text
1:08 rather than a voice or phone interface
1:11 they may check status of customer orders
1:13 keep records and directions and assess
1:15 the accuracy of customer complaints and
1:18 if you work in a company with say many
1:19 customer service Representatives the
1:22 first step to analyzing the potential
1:24 for using gender AI would be to
1:26 understand for your business what are
1:29 the tasks that the representatives in
1:32 your company do after that we can then
1:35 take a look at these different tasks and
1:38 try to assess their potential for
1:41 generative AI to either hope or augment
1:43 or to automate these thoughts for
1:45 example for generative AI to pick up the
1:47 phone and have a long conversation
1:49 that's still pretty difficult so we
1:52 assess that to be a lower potential
1:55 opportunity but answering text chat with
1:57 customers that might have a higher
1:59 potential maybe checking status of
2:01 customer orders is medium whereas
2:04 keeping records of customer interactions
2:06 could be high and assessing accuracy of
2:08 customer complaints may be low all of
2:11 these examples in the rightmost column
2:13 are hypothetical and the actual impact
2:15 on your business will be different and
2:17 would depend on the specifics of your
2:19 business but after an analysis like this
2:21 and I'll go in a little bit into the
2:23 specifics of how to carry out this
2:25 analysis you might then decide that
2:27 answering customer chat queries and
2:29 keeping records of customer interactions
2:31 have The Highest Potential and therefore
2:34 Focus your efforts on those two tasks
2:37 now the opportunity for Gen of AI could
2:41 be either augmentation or automation by
2:44 augmentation I mean we can use AI to
2:46 help a human with a Tas in the customer
2:48 service representative context we might
2:51 have genbi recommended response for a
2:52 customer service agent to edit or
2:54 approve but not fully automate the
2:56 sending of a message back to the
2:58 customer so if we're not sure yet if the
3:01 Gent of AI would give good answers then
3:04 recommending a response could speed up
3:06 the people doing the work but not fully
3:08 automated and this would be an example
3:12 of augmentation and automation would be
3:14 if we have an AI system fully
3:17 automatically perform at all so if we
3:18 were to automatically transcribe and
3:20 summarize records of customer
3:22 interactions um that could be an example
3:24 of automation what I see in many
3:27 applications is that businesses will
3:29 sometimes start with augmentation to
3:31 maybe let a human double check or
3:35 finalize the output before this use but
3:36 then as you gain trust and gain
3:38 confidence in the output of the Gena of
3:41 AI then the user interface can be
3:43 adapted to make the process more and
3:46 more efficient for humans and to then
3:48 gradually shift toward higher and higher
3:50 degrees of augmentation and perhaps
3:54 eventually to full automation now given
3:56 a list of tasks like this how do you
3:58 come up with this column on the right
4:00 how do you evaluate the different tasks
4:04 for G AI potential the potential for
4:06 augmenting or automating a task depends
4:08 mostly on two things technical
4:11 feasibility and business value so
4:14 technical feasibility refers to can AI
4:17 do it and also how costly is it to build
4:21 an AI system to do it and with regard to
4:24 using an LM I found the framework we
4:26 discussed last week of asking can a
4:28 fresh college graduate following the
4:31 instructions in the prompt the task that
4:33 could give you a first guess an
4:36 imperfect not necessary fully accurate
4:38 guess but it gives you a way to think
4:40 about whether a certain task may be
4:43 doable or not and sometimes if you're
4:46 not sure if an LM can do a certain task
4:48 I would encourage you to try prompting
4:51 an LM to see if you can get the LM to do
4:52 that task and this would be an
4:54 experiment that you might be able to do
4:56 quite quickly so long as you're not
4:59 revealing confidential information if
5:02 you you take some prompts for say
5:04 answering customer chat queries and
5:07 paste them into a large language model
5:10 you can maybe quickly get a sense of how
5:13 good RS response is and this could help
5:15 you relatively quickly assess technical
5:18 feasibility of using generative AI for a
5:20 particular cost and an AI engineer can
5:23 also help you assess if more Advanced
5:25 Techniques like rack retrieve alment
5:27 generation fine-tuning or other
5:30 techniques can help and also give you a
5:33 sense of perhaps the complexity and
5:35 therefore the cost of building an AI
5:38 system to tackle a certain task in this
5:40 course I'm focusing mainly on technical
5:44 feasibility using gen of AI technology
5:46 if you or your team is familiar with
5:48 other AI tools such as supervised
5:51 learning you can also assess the
5:53 technical feasibility of using other
5:56 tools as well for augmenting or
5:58 automating different tasks other than
6:00 technical feasibility the second
6:02 criteria I urge you to think through is
6:05 the business value so how valuable is it
6:08 to use AI to either augment or automate
6:11 a particular task and so the questions I
6:14 would ask to frame up my thinking on
6:16 business value would be things like how
6:18 much time is spent on this task so how
6:19 much time savings can we actually
6:22 realize second I'd also ask does during
6:25 this task significantly faster cheaper
6:27 or more consistently using AI create
6:31 substantial value while it may seem like
6:33 augmentation and automation helps lead
6:36 to cost savings we'll see later this
6:38 week as well that when you automate a
6:40 TSS sometimes the benefits are much
6:43 greater than just cost savings because
6:46 it also lead to rethinking the workflow
6:50 around that toss but if what I'm saying
6:51 doesn't make sense yet don't worry about
6:53 it we'll see some specific examples of
6:55 this later this week before wrap up this
6:57 video there's one more resource I want
7:00 to share that maybe us useful for your
7:02 analysis of how to break job RS down
7:05 into task which is that there are online
7:08 occupation databases that you can look
7:11 up to see what are the tasks that
7:13 comprise a certain roow here's a
7:15 screenshot from a website called onet
7:19 which is a US government funded website
7:20 that for the customer service
7:22 representative row lists lots of
7:24 different tasks including confir of
7:26 customers by telephone or in person key
7:28 records of customer interactions and so
7:30 on a found that occupation databases
7:33 like this tend to be General and not
7:36 necessarily specific to your company and
7:38 so I wouldn't recommend just using the
7:41 results from say this onet database and
7:43 assuming it's accurate for your company
7:45 there'll usually be some entries there
7:47 that you read and feel like no this
7:49 doesn't seem like it applies to my
7:51 company but I found that this is useful
7:52 resource to take a look at just for
7:55 ideas and to help make sure that maybe
7:56 you haven't missed anything when
7:58 thinking through what are the tasks done
8:00 by people in different job roles in your
8:04 company onet is a little bit us Centric
8:07 but has a nice easy to use user
8:08 interface so that encourage you to play
8:10 with it and there are some other
8:12 countries as well that have some other
8:14 country or region specific databases
8:16 that you may able to find online as well
8:19 but I found that for many job RS onet is
8:22 may be a reasonable initial starting
8:24 point so that's how you can look at
8:26 different job roles and start to break
8:29 them down in the TSS and analyze the IND
8:31 indidual task for potential for
8:34 augmentation or Automation and I hope
8:37 you play with the onet website and get a
8:39 feel for what different tasks and
8:41 different job roles look like in this
8:43 video we went through the customer
8:45 service representative example I'd like
8:48 to go through you a few examples of
8:50 other job roles as well so let's go take
