Can we personalise learning at scale? with Ashish Bansal (#367)
What if the most useful AI tutor is the one that refuses to give your child the answer?
In this episode, Ben and Steve sit down with Ashish Bansal, co-founder and CEO of StarSpark AI and former Google, Twitter and Amazon engineer, to ask what AI tutoring should really look like when learning, not engagement, is the goal. From his son gaming an online algebra course by memorising a seven-question bank, to why most AI chatbots are built for the wrong purpose entirely, Ashish makes the case for purpose-built, Socratic AI that makes students think harder, not less.
Ashish Bansal is the co-founder and CEO of StarSpark AI, an affordable Socratic AI maths tutor on a mission to bring a capable tutor to every student in the world. A specialist in large language models and recommendation systems, he previously worked at Google, including with the Gemini team, as well as at Twitter and Twitch. Before building StarSpark, he completed teacher training at Stanford to learn how teachers teach. He is also a dad of two, and it was watching his own children learn that sparked the company.
We cover:
- Why "solving is not tutoring" and great mathematicians don't automatically make great teachers
- What parents and technologists misunderstand about pedagogy and age-appropriate learning
- How StarSpark refuses to be an answer machine and uses Socratic questioning instead
- Sequencing question difficulty using Webb's Depth of Knowledge
- Why multiple choice encourages guessing, and how his son gamed an online algebra course
- The 80% problem: how learning gaps compound year after year in maths
- Personalised learning versus group instruction, and Bloom's two sigma problem
- Socrates, gurukuls and why our school systems are built around a cost model
- Flipping the classroom: AI teaches the concept, the teacher helps when students get stuck
- Why the school year is still nine months, and whether learning should be continuous
- Why general-purpose chatbots are built for engagement, not learning outcomes
- Purpose-built AI vs wrappers: grounding in textbooks, computer algebra systems and avoiding emotional dependence
- The Star Wars role-play that got one student through 119 maths problems in two days
- Practical things teachers can do with AI tomorrow to give every student more individual attention
Why you should listen
If you're a teacher, school leader or parent trying to work out whether AI will help young people learn or simply do their thinking for them, this conversation offers a rare inside view from someone who builds the technology and still puts pedagogy first. You'll come away with sharper questions to ask of any AI product and practical ideas you can try in your classroom straight away.
Chapters
00:00 Highlights
01:01 Welcome to the Edufuturists podcast
02:17 Community updates and XP School
04:08 Meet Ashish Bansal: from Google to StarSpark AI
05:33 How being a dad fuelled the mission
06:43 Why solving is not tutoring
09:02 What parents and technologists miss about pedagogy
11:25 Lessons from Google, Twitter and Twitch
14:08 Is AI stopping students from thinking?
15:50 No answer machine: building a Socratic tutor
17:41 Sequencing difficulty with Webb's Depth of Knowledge
18:41 Why multiple choice fails, and gaming Khan Academy
19:56 The fastest route vs the learning process
22:24 What does it actually mean to learn?
23:36 Personalised learning vs group instruction
24:29 AD: Brilliant Festival
25:33 The 80% problem: how maths gaps compound
26:59 Is personalisation the future of education?
27:50 Socrates, gurukuls and the cost model of schooling
29:50 Bloom's two sigma, Biesta and learning in community
32:18 The future of education: humans and AI together
35:45 Flipping the model: AI teaches, teachers unstick
36:50 Mastery, grade boundaries and the nine-month school year
38:40 AI adoption in schools: square peg, round hole
40:50 Chatbots built for engagement, not learning
43:14 Purpose-built AI vs wrappers
44:30 Not your friend: guardrails and the Star Wars story
47:02 What teachers can do with AI tomorrow
50:16 Quick-fire questions
51:58 Wrap up and thanks
Thanks so much for joining us again for another episode - we appreciate you.
Ben & Steve x
https://www.brilliantfestival.com/?utm_source=social&utm_medium=partners&utm_campaign=edufuturists
Check out all about Edufuturists at https://www.edufuturists.com
Want to sponsor future episodes or get involved with the Edufuturists work?
Get in touch: info@edufuturists.com
Uprising tickets are LIVE: https://www.edufuturists.com/uprising-awards/uprising2027
Grab your copy of the new Pick 'n' Mix Education book at: https://www.crownhouse.co.uk/picknmix-education
Join the Wolfpack: edufuturists.com/wolfpack
What if the most useful AI tutor is the one that refuses to give your child the answer?
In this episode, Ben and Steve sit down with Ashish Bansal, co-founder and CEO of StarSpark AI and former Google, Twitter and Amazon engineer, to ask what AI tutoring should really look like when learning, not engagement, is the goal. From his son gaming an online algebra course by memorising a seven-question bank, to why most AI chatbots are built for the wrong purpose entirely, Ashish makes the case for purpose-built, Socratic AI that makes students think harder, not less.
Ashish Bansal is the co-founder and CEO of StarSpark AI, an affordable Socratic AI maths tutor on a mission to bring a capable tutor to every student in the world. A specialist in large language models and recommendation systems, he previously worked at Google, including with the Gemini team, as well as at Twitter and Twitch. Before building StarSpark, he completed teacher training at Stanford to learn how teachers teach. He is also a dad of two, and it was watching his own children learn that sparked the company.
We cover:
- Why "solving is not tutoring" and great mathematicians don't automatically make great teachers
- What parents and technologists misunderstand about pedagogy and age-appropriate learning
- How StarSpark refuses to be an answer machine and uses Socratic questioning instead
- Sequencing question difficulty using Webb's Depth of Knowledge
- Why multiple choice encourages guessing, and how his son gamed an online algebra course
- The 80% problem: how learning gaps compound year after year in maths
- Personalised learning versus group instruction, and Bloom's two sigma problem
- Socrates, gurukuls and why our school systems are built around a cost model
- Flipping the classroom: AI teaches the concept, the teacher helps when students get stuck
- Why the school year is still nine months, and whether learning should be continuous
- Why general-purpose chatbots are built for engagement, not learning outcomes
- Purpose-built AI vs wrappers: grounding in textbooks, computer algebra systems and avoiding emotional dependence
- The Star Wars role-play that got one student through 119 maths problems in two days
- Practical things teachers can do with AI tomorrow to give every student more individual attention
Why you should listen
If you're a teacher, school leader or parent trying to work out whether AI will help young people learn or simply do their thinking for them, this conversation offers a rare inside view from someone who builds the technology and still puts pedagogy first. You'll come away with sharper questions to ask of any AI product and practical ideas you can try in your classroom straight away.
Chapters
00:00 Highlights
01:01 Welcome to the Edufuturists podcast
02:17 Community updates and XP School
04:08 Meet Ashish Bansal: from Google to StarSpark AI
05:33 How being a dad fuelled the mission
06:43 Why solving is not tutoring
09:02 What parents and technologists miss about pedagogy
11:25 Lessons from Google, Twitter and Twitch
14:08 Is AI stopping students from thinking?
15:50 No answer machine: building a Socratic tutor
17:41 Sequencing difficulty with Webb's Depth of Knowledge
18:41 Why multiple choice fails, and gaming Khan Academy
19:56 The fastest route vs the learning process
22:24 What does it actually mean to learn?
23:36 Personalised learning vs group instruction
24:29 AD: Brilliant Festival
25:33 The 80% problem: how maths gaps compound
26:59 Is personalisation the future of education?
27:50 Socrates, gurukuls and the cost model of schooling
29:50 Bloom's two sigma, Biesta and learning in community
32:18 The future of education: humans and AI together
35:45 Flipping the model: AI teaches, teachers unstick
36:50 Mastery, grade boundaries and the nine-month school year
38:40 AI adoption in schools: square peg, round hole
40:50 Chatbots built for engagement, not learning
43:14 Purpose-built AI vs wrappers
44:30 Not your friend: guardrails and the Star Wars story
47:02 What teachers can do with AI tomorrow
50:16 Quick-fire questions
51:58 Wrap up and thanks
Thanks so much for joining us again for another episode - we appreciate you.
Ben & Steve x
https://www.brilliantfestival.com/?utm_source=social&utm_medium=partners&utm_campaign=edufuturists
Check out all about Edufuturists at https://www.edufuturists.com
Want to sponsor future episodes or get involved with the Edufuturists work?
Get in touch: info@edufuturists.com
Uprising tickets are LIVE: https://www.edufuturists.com/uprising-awards/uprising2027
Grab your copy of the new Pick 'n' Mix Education book at: https://www.crownhouse.co.uk/picknmix-education
Join the Wolfpack: edufuturists.com/wolfpack
Explore More
