Connected　and　Automated　Vehicles　(CAVs)　will　revolutionize　the　future　intelligent　transportation　system　by　enabling　fully　automated　vehicular　control　such　as　accelerating,　steering　and　lane　changes,　etc.　Among　them,　lane　changes　are　vital　aspects　since　they　require　not　only　local　CAV　control,　but　also　interactions　with　surrounding　CAVs　to　guarantee　the　overall　traffic　efficiency.　Nevertheless,　the　Line-of-Sight　(LoS)　sensing　range　adds　significant　limitations　on　CAVs　to　efficiently　make　lane　change　decisions.　Moreover,　the　self-interested　CAV　lane　change　decision　ignores　its　impact　on　the　surrounding　CAVs　as　well　as　the　entire　traffic　flow.　In　this　article,　we　propose　an　edge　Artificial　intelligence　(AI)　empowered　CAV　framework　to　enhance　the　sensing　and　decision　making　abilities　of　CAVs　and　thus　assist　in　efficient　lane　changes　with　foresighted　and　cooperative　vehicular　intelligence.　Under　the　proposed　framework,　NonLine-of-Sight　(NLoS)　lane　environment　data　can　be　collected　by　the　edge　AI　node　and　aggregated　with　the　LoS　data　collected　by　the　onboard　CAV　sensors.　The　aggregated　data　can　be　utilized　to　train　a　deep　reinforcement　learning　model　that　enables　CAVs　to　make　foresighted　lane　change　decisions　by　learning　from　the　NLoS　environment　data.　Besides,　we　incorporate　cooperative　game　theory　with　the　deep　reinforcement　learning　model　to　help　CAVs　evaluate　their　impact　on　the　surrounding　CAVs　and　make　cooperative　lane　change　decisions　to　improve　the　overall　traffic　efficiency.　Simulations　are　conducted　to　prove　the　effectiveness　of　the　proposed　edge　AI　empowered　CAV　lane　changes.　©　2024　IEEE.